NavAble commited on
Commit
ece131e
·
verified ·
1 Parent(s): 7f7d58f

Dataset card: paper, authors, tools and figures

Browse files
README.md CHANGED
@@ -1,11 +1,12 @@
1
  ---
2
- pretty_name: "BLV Object Recognition (Synthetic + Real-World)"
3
  license: cc-by-4.0
4
  language:
5
  - en
6
  task_categories:
7
  - image-segmentation
8
  - object-detection
 
9
  task_ids:
10
  - semantic-segmentation
11
  - instance-segmentation
@@ -14,9 +15,13 @@ size_categories:
14
  tags:
15
  - accessibility
16
  - blind-low-vision
 
 
 
17
  - synthetic
18
  - isaacsim
19
  - sim-to-real
 
20
  - urban-navigation
21
  configs:
22
  - config_name: syn
@@ -123,244 +128,444 @@ dataset_info:
123
  dtype: int32
124
  ---
125
 
126
- # BLV Object Recognition: Synthetic + Real-World
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
127
 
128
- A dataset for training and evaluating object recognition and segmentation
129
- models on infrastructure relevant to **blind and low-vision (BLV) navigation**
130
- in urban environments. Three configurations plus a flat tree of 3D assets:
131
 
132
- | Config / tree | Splits | Purpose |
133
- |---------------|--------|---------|
134
- | `syn` | `train` | Photorealistic IsaacSim renders with depth, surface normals and 2D/3D boxes, for training / pretraining. |
135
- | `real_ours` | `train` / `validation` / `test` | Real photographs we captured. **`real_ours/test` is the canonical benchmark eval.** |
136
- | `real_curated` | `train` | Curated frames from public HF segmentation datasets (`curation`, `mapillary`), remapped to our class palette. |
137
- | `synthetic_objects/` (tree) | n/a | 3D asset library: per-asset `.glb` + `.ply` + `.usdz` triples grouped by BLV class. |
138
 
139
- ![samples](docs/hero.png)
140
 
141
- ## Quick links
 
 
142
 
143
- - [Datasheet for Datasets](docs/datasheet.md)
144
- - [Class index + palette](class_index.json)
145
- - Croissant metadata is auto-generated by Hugging Face for this repo (look for the *Croissant* button on the dataset page).
146
- - Paper: NeurIPS 2026 Datasets & Benchmarks (TBD).
 
147
 
148
- ## Loading
 
 
 
149
 
150
- ### With `datasets`
151
 
152
- ```python
153
- from datasets import load_dataset
154
 
155
- syn_train = load_dataset("NavAble/NeurIPS_2026_BLV", "syn", split="train")
156
- ours_train = load_dataset("NavAble/NeurIPS_2026_BLV", "real_ours", split="train")
157
- ours_val = load_dataset("NavAble/NeurIPS_2026_BLV", "real_ours", split="validation")
158
- ours_test = load_dataset("NavAble/NeurIPS_2026_BLV", "real_ours", split="test") # canonical eval
159
- curated_train = load_dataset("NavAble/NeurIPS_2026_BLV", "real_curated", split="train")
160
 
161
- row = ours_test[0]
162
- row["image"] # PIL.Image.Image, RGB
163
- row["mask"] # PIL.Image.Image, P-mode (palette) - pixel value == class_id
164
- ```
165
 
166
- ### Pulling the 3D assets
 
 
167
 
168
  ```python
169
- from huggingface_hub import snapshot_download
170
-
171
- # All 3D assets for a single class:
172
- snapshot_download(
173
- repo_id="NavAble/NeurIPS_2026_BLV", repo_type="dataset",
174
- allow_patterns=["synthetic_objects/door_button/**"],
175
- local_dir="./assets",
176
- )
177
  ```
178
 
179
- ### With PyTorch directly
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
180
 
181
  ```python
182
- from torch.utils.data import Dataset
183
- from datasets import load_dataset
184
- import numpy as np
185
- import torch
186
- import torchvision.transforms.functional as TF
187
-
188
- class BLVSegDataset(Dataset):
189
- def __init__(self, config: str, split: str, image_size: int = 512):
190
- self.ds = load_dataset("NavAble/NeurIPS_2026_BLV", config, split=split)
191
- self.image_size = image_size
192
-
193
- def __len__(self):
194
- return len(self.ds)
195
-
196
- def __getitem__(self, idx):
197
- row = self.ds[idx]
198
- img = TF.resize(row["image"].convert("RGB"), [self.image_size, self.image_size])
199
- mask = TF.resize(row["mask"], [self.image_size, self.image_size],
200
- interpolation=TF.InterpolationMode.NEAREST)
201
- img_t = TF.to_tensor(img)
202
- mask_t = torch.from_numpy(np.array(mask, dtype=np.int64))
203
- return {"image": img_t, "mask": mask_t, "class": row["object_class"]}
204
  ```
205
 
206
- ## Splits & sizes
207
 
208
- | Config | Split | Rows |
209
- |----------------|--------------|------|
210
- | `syn` | train | 450860 |
211
- | `real_ours` | train | 3703 |
212
- | `real_ours` | validation | 396 |
213
- | `real_ours` | test | 1482 |
214
- | `real_curated` | train | 36466 |
215
 
216
- 3D asset library (`synthetic_objects/`): 500 GLB+PLY+USDZ triples across 9 classes.
217
 
218
- ## Class taxonomy
 
 
219
 
220
- | ID | Class | Synthetic | Real (Ours) |
221
- |----|-------|-----------|-------------|
222
- | 1 | `aps_button` | yes | yes |
223
- | 2 | `bus_stop` | yes | yes |
224
- | 3 | `bus_stop_sign` | yes | yes |
225
- | 4 | `crosswalk` | yes | yes |
226
- | 5 | `door_button` | yes | yes |
227
- | 6 | `elevator` | yes | yes |
228
- | 7 | `elevator_button` | yes | yes |
229
- | 8 | `escalator` | yes | yes |
230
- | 9 | `handrail` | yes | yes |
231
- | 10 | `pedestrian_signal` | yes | yes |
232
- | 11 | `revolving_door` | yes | no |
233
 
234
- The synthetic-only class `revolving_door` has no real-world examples in this release;
235
- report real-world metrics over the 10 shared classes.
 
 
 
 
236
 
237
- ## Per-class row counts
 
238
 
239
- | Class | syn/train | real_ours/train | real_ours/val | real_ours/test | real_curated/train |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
240
  |---|---|---|---|---|---|
241
- | `aps_button` | 62690 | 206 | 23 | 66 | 0 |
242
- | `bus_stop` | 60770 | 205 | 23 | 62 | 0 |
243
- | `bus_stop_sign` | 60450 | 140 | 16 | 54 | 0 |
244
- | `crosswalk` | 53970 | 9 | 1 | 3 | 27786 |
245
- | `door_button` | 45360 | 1327 | 148 | 622 | 0 |
246
- | `elevator` | 23350 | 1065 | 119 | 479 | 15 |
247
- | `elevator_button` | 23350 | 378 | 23 | 86 | 4401 |
248
- | `escalator` | 6420 | 135 | 15 | 40 | 1296 |
249
- | `handrail` | 44840 | 21 | 3 | 8 | 1197 |
250
- | `pedestrian_signal` | 44650 | 217 | 25 | 62 | 6650 |
251
- | `revolving_door` | 25010 | 0 | 0 | 0 | 0 |
252
-
253
- ## Synthetic coverage
254
-
255
- - **Frames:** 450860
256
- - **Trajectories:** 700
257
- - **Environments:** 37
258
- - **Distinct assets:** 110
259
-
260
- ## Mask encoding
261
-
262
- Each `mask` is a single-channel PNG (PIL `mode="P"`) with an embedded palette.
263
- Pixel value `i` corresponds to the `i`-th entry in `class_index.json`:
264
-
265
- | Pixel | Class | Palette RGB |
266
- |-------|-------|-------------|
267
- | 0 | `BACKGROUND` | (0, 0, 0) |
268
- | 1 | `aps_button` | (220, 20, 60) |
269
- | 2 | `bus_stop` | (255, 140, 0) |
270
- | 3 | `bus_stop_sign` | (255, 215, 0) |
271
- | 4 | `crosswalk` | (50, 205, 50) |
272
- | 5 | `door_button` | (0, 191, 255) |
273
- | 6 | `elevator` | (138, 43, 226) |
274
- | 7 | `elevator_button` | (255, 105, 180) |
275
- | 8 | `escalator` | (0, 128, 128) |
276
- | 9 | `handrail` | (165, 42, 42) |
277
- | 10 | `pedestrian_signal` | (75, 0, 130) |
278
- | 11 | `revolving_door` | (255, 20, 147) |
279
-
280
- Convert to a numeric label map with `np.array(row["mask"])`.
281
-
282
- ## Synthetic modalities (`syn`)
283
-
284
- | Column | Content |
285
- |--------|---------|
286
- | `image` | RGBA render, 1280x720 PNG |
287
- | `mask` | class mask, palettized PNG (see *Mask encoding*) |
288
- | `normals` | surface normals, RGBA PNG as rendered by Replicator |
289
- | `depth` | distance to the image plane, float16 `.npy` bytes, 720x1280, in metres; `inf` where nothing was hit |
290
- | `bbox`, `bbox_class_ids`, `occlusion_ratio`, `bbox_prim_paths` | 2D tight boxes `[x_min, y_min, x_max, y_max]` (pixels), class ids, occlusion and USD prim paths |
291
- | `bbox_3d`, `bbox_3d_transform`, `bbox_3d_class_ids`, `bbox_3d_occlusion_ratio`, `bbox_3d_prim_paths` | 3D boxes: local extents `[x_min, y_min, z_min, x_max, y_max, z_max]`, row-major 4x4 local-to-world transform (translation in the last row, world units are centimetres, as in the IsaacSim stage), class ids, occlusion and prim paths |
292
 
293
  ```python
294
- import io, numpy as np
295
- row = syn_train[0]
296
- depth = np.load(io.BytesIO(row["depth"])) # float16 (720, 1280)
297
- normals = row["normals"] # PIL.Image, RGBA
298
  ```
299
 
300
- ## Source data
301
-
302
- - **Synthetic** — generated in NVIDIA IsaacSim (Replicator) by spawning each
303
- asset across a curated catalog of urban environments
304
- (37 unique scenes including default and sunset/night lighting
305
- variants), with randomized camera trajectories. Each frame ships an RGBA
306
- image, a semantic class mask, a surface-normal map, a depth map
307
- (distance to the image plane), and 2D tight and 3D bounding boxes with
308
- their USD prim paths.
309
- - **Real (Ours)** — real photographs captured at 113 distinct
310
- physical locations covering the 10 shared object classes. Annotations were
311
- authored as polygon segmentations in COCO format and rasterized to the
312
- unified palettized PNG mask format used here.
313
- - **Real (Curated)** — frames sampled from public segmentation datasets
314
- (`source_dataset = "curation"` or `"mapillary"`). Class IDs were remapped
315
- from each source taxonomy to the global BLV palette. The original
316
- per-frame split (`split_origin`) is preserved as a column; all curated
317
- rows are exposed under a single `train` split here.
318
- - **3D Assets (`synthetic_objects/`)** — 500 per-asset folders, each
319
- containing a `.glb` (Khronos), a Gaussian-splat `.ply`, and a `.usdz`
320
- (USD bundle, Apple/Pixar) ready for IsaacSim or AR pipelines. Assets are
321
- organized by BLV class.
322
-
323
- ## Preprocessing
324
-
325
- Produced by `scripts/build_hf_dataset.py`. Synthetic RGB PNGs are hardlinked
326
- unchanged from the source tree; the IsaacSim RGBA-encoded semantic masks are
327
- converted into single-channel palettized PNGs against a global class index;
328
- synthetic 2D bounding-box `.npy` files are flattened into JSONL columns; the
329
- real-world COCO polygon annotations are rasterized to the same palettized PNG
330
- format using `pycocotools`. For `syn`, surface normals are stored as the rendered PNGs, depth is converted from float32 to float16 (`.npy` bytes; at most 0.05% relative error), and 3D boxes keep their local extents and 4x4 transforms.
331
-
332
- ## Known limitations
333
-
334
- - **Resolution mismatch.** Synthetic frames are 1280×720; real-world frames
335
- are 640×360. Models that resize to a common input shape are unaffected.
336
- - **Class imbalance in real-world data.** Some classes have few real-world
337
- examples (e.g. `crosswalk`, `handrail`). Report per-class mIoU alongside any
338
- aggregate.
339
- - **`revolving_door` is synthetic-only.** Evaluate over the 10 shared classes for
340
- real-world metrics.
341
- - **Sim-to-real gap.** Synthetic textures and lighting may not match
342
- real-world distributions perfectly.
343
-
344
- ## Ethical considerations
345
-
346
- - The synthetic data contains no personally identifiable information.
347
- - Real-world captures were collected in public spaces; the dataset is intended
348
- for accessibility research and **must not be used for surveillance or
349
- identification of individuals**.
350
- - The class taxonomy targets infrastructure relevant to blind/low-vision
351
- navigation; models trained on this dataset should not be deployed in
352
- safety-critical settings without additional validation.
353
 
354
- ## Changelog
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
355
 
356
- - **2026-11**: `syn` re-rendered and re-released with surface normals, depth and 3D
357
- bounding boxes (450,860 frames). Class 11 is now named `revolving_door` (same id and
358
- color; its 3D assets moved to `synthetic_objects/revolving_door/`). Other configs are
359
- unchanged.
360
 
361
- ## License
 
 
 
 
 
 
 
 
362
 
363
- Released under **CC BY 4.0**.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
364
 
365
  ## Citation
366
 
@@ -376,3 +581,10 @@ Released under **CC BY 4.0**.
376
  url = {https://neurips.cc/virtual/2026/loc/sydney/poster/138983}
377
  }
378
  ```
 
 
 
 
 
 
 
 
1
  ---
2
+ pretty_name: "NavAble: Accessibility-Critical Object Recognition for Blind Navigation"
3
  license: cc-by-4.0
4
  language:
5
  - en
6
  task_categories:
7
  - image-segmentation
8
  - object-detection
9
+ - depth-estimation
10
  task_ids:
11
  - semantic-segmentation
12
  - instance-segmentation
 
15
  tags:
16
  - accessibility
17
  - blind-low-vision
18
+ - assistive-technology
19
+ - robotics
20
+ - navigation
21
  - synthetic
22
  - isaacsim
23
  - sim-to-real
24
+ - 3d-assets
25
  - urban-navigation
26
  configs:
27
  - config_name: syn
 
128
  dtype: int32
129
  ---
130
 
131
+ # NavAble: A Large-Scale Dataset and Synthetic Data Generation Pipeline for Blind Navigation
132
+
133
+ <p align="center">
134
+ <a href="https://arxiv.org/abs/2610.19330"><img alt="arXiv" src="https://img.shields.io/badge/arXiv-2610.19330-b31b1b.svg"></a>
135
+ <a href="https://guidedogrobot-navable.github.io/"><img alt="Project page" src="https://img.shields.io/badge/Project-Page-1f6feb.svg"></a>
136
+ <a href="https://github.com/DARoSLab/NavAble"><img alt="Code" src="https://img.shields.io/badge/GitHub-Code-181717.svg?logo=github"></a>
137
+ <img alt="NeurIPS 2026" src="https://img.shields.io/badge/NeurIPS-2026-8a2be2.svg">
138
+ <img alt="License" src="https://img.shields.io/badge/License-CC%20BY%204.0-lightgrey.svg">
139
+ </p>
140
+
141
+ <p align="center">
142
+ <a href="https://hchlhwang.github.io/">Hochul Hwang</a><sup>†</sup>,
143
+ <a href="https://jahirsadik.github.io/">Jahir Sadik Monon</a><sup>†</sup>,
144
+ <a href="https://www.linkedin.com/in/soowan-yang-0b723a293/">Soowan Yang</a>,
145
+ <a href="https://www.shivenpatel.me/">Shiven Umeshbhai Patel</a>,
146
+ <a href="https://www.linkedin.com/in/anhnnguyennh/">Anh Nhat Hong Nguyen</a>,
147
+ <a href="https://www.linkedin.com/in/kien-nguyen-umass/">Kien Nguyen</a>,
148
+ <a href="https://www.linkedin.com/in/keshavgargcs/">Keshav Garg</a>,
149
+ <a href="https://www.linkedin.com/in/dylan-gage-4b1107331/">Dylan Gage</a>,
150
+ <a href="https://www.linkedin.com/in/duretti-hordofaa/">Duretti Hordofaa</a>,
151
+ <a href="https://www.linkedin.com/in/anshu-anjna-08a651326/">Anshu Anjna</a>,
152
+ <a href="https://eshed1.github.io/">Eshed Ohn-Bar</a>,
153
+ <a href="https://www.umass.edu/robotics/people/donghyun-kim">Donghyun Kim</a>
154
+ <br>
155
+ University of Massachusetts Amherst · DGIST · Boston University &nbsp;&nbsp; <sup>†</sup>Equal contribution
156
+ <br>
157
+ <b>NeurIPS 2026</b>
158
+ </p>
159
+
160
+ NavAble helps vision models find the things blind and low-vision (BLV) travelers depend on:
161
+ accessible pedestrian signal (APS) buttons, door buttons, pedestrian signals, handrails, elevators,
162
+ escalators, bus stops and more. These objects are rare or inconsistently labelled in existing vision
163
+ datasets. NavAble brings together:
164
+
165
+ - **NavAble-Real**: 5,581 newly collected photos with dense class masks, including a location-disjoint **benchmark test set of 1,482 images**;
166
+ - **NavAble-Curated**: 36,466 images from public datasets, curated and remapped to one 11-class taxonomy;
167
+ - **NavAble-Synthetic**: 450,860 photorealistic NVIDIA Isaac Sim frames with class masks, **depth, surface normals and 2D/3D boxes**;
168
+ - the **500 3D assets** used to render them, and **28 trained segmentation checkpoints** from the paper.
169
+
170
+ ![NavAble samples](docs/teaser.png)
171
+
172
+ **Contents:** [News](#news) · [At a glance](#at-a-glance) · [Quick start](#quick-start) · [Dataset structure](#dataset-structure) · [How to use](#how-to-use) · [Results](#results-from-the-paper) · [How the data was made](#how-the-data-was-made) · [Tools](#tools) · [Dataset card](#dataset-card) · [Citation](#citation)
173
+
174
+ ## News
175
+
176
+ - **2026-11**: `syn` re-rendered and re-released with surface normals, depth and 3D bounding boxes (450,860 frames). Class 11 is now named `revolving_door` (same id and color; its 3D assets moved to `synthetic_objects/revolving_door/`). New [`tools/`](tools/) for downloading, exploring, exporting and evaluating.
177
+ - **2026**: The paper is accepted at NeurIPS 2026. Preprint: [arXiv:2610.19330](https://arxiv.org/abs/2610.19330).
178
+
179
+ ## At a glance
180
+
181
+ | Subset | Config / path | Splits (rows) | Size | Image size | Annotations |
182
+ |---|---|---|---|---|---|
183
+ | **Real, collected** | `real_ours` | train 3,703 · validation 396 · **test 1,482** | 1.5 GB | 640×360 (a few 640×480) | class masks, capture location id |
184
+ | **Real, curated** | `real_curated` | train 36,466 | 114 GB | mostly 640×640 | class masks remapped from the source datasets |
185
+ | **Synthetic** | `syn` | train 450,860 | 816 GB | 1280×720 | class masks, depth, surface normals, 2D and 3D boxes, USD prim paths |
186
+ | **3D assets** | `synthetic_objects/` | 500 assets in 9 classes | 23 GB | – | `.glb` mesh, `.ply` Gaussian splat, `.usdz` |
187
+ | **Checkpoints** | `navable_*.pth` | 7 models × 4 training-data mixes | 31 GB | – | semantic segmentation models |
188
 
189
+ - **11 classes**; 10 of them have real-world examples, and `revolving_door` is synthetic-only.
190
+ - Real photos come from 113 physical locations. The **test split shares no location** with train or validation.
191
+ - Synthetic frames cover 37 environments (with day, sunset and night variants), 110 distinct assets and 25,350 rendered camera trajectories.
192
 
193
+ ## Quick start
 
 
 
 
 
194
 
195
+ ### Load with 🤗 Datasets
196
 
197
+ ```python
198
+ from datasets import load_dataset
199
+ import numpy as np
200
 
201
+ test = load_dataset("NavAble/NeurIPS_2026_BLV", "real_ours", split="test") # fetches real_ours (1.5 GB)
202
+ row = test[0]
203
+ image = row["image"] # PIL image, RGB
204
+ labels = np.array(row["mask"]) # (H, W) uint8; pixel value = class id, 0 = background
205
+ print(row["object_class"], row["location"], np.unique(labels))
206
 
207
+ # syn is 816 GB: stream it rather than downloading it
208
+ syn = load_dataset("NavAble/NeurIPS_2026_BLV", "syn", split="train", streaming=True)
209
+ frame = next(iter(syn)) # image, mask, normals, depth, boxes, ...
210
+ ```
211
 
212
+ Loading any split of a config downloads that whole config. Streaming (`streaming=True`) reads rows on demand instead.
213
 
214
+ ### Get the tools
 
215
 
216
+ The [`tools/`](tools/) folder has a small helper module and command-line scripts. They read only the
217
+ parquet shards they need.
 
 
 
218
 
219
+ ```bash
220
+ pip install -U datasets huggingface_hub pyarrow pillow numpy
221
+ hf download NavAble/NeurIPS_2026_BLV --repo-type dataset --include "tools/*" --local-dir navable
222
+ cd navable
223
 
224
+ python tools/explore.py --config syn --classes escalator handrail --num 4 --out syn.png # look at samples
225
+ python tools/download.py --real --syn escalator --dry-run # plan a partial download
226
+ ```
227
 
228
  ```python
229
+ import sys; sys.path.insert(0, "tools")
230
+ import navable as nv
231
+
232
+ escalators = nv.load_syn(["escalator"]) # streams only the 8 escalator shards (12 GB, not 816 GB)
233
+ row = next(iter(escalators))
234
+ depth = nv.decode_depth(row["depth"]) # float16 (720, 1280), metres
235
+ nv.overlay(row["image"], nv.mask_array(row["mask"])).save("overlay.png")
 
236
  ```
237
 
238
+ ## Dataset structure
239
+
240
+ ### Repository layout
241
+
242
+ ```
243
+ NavAble/NeurIPS_2026_BLV
244
+ ├── README.md this card (configs and features in the YAML header)
245
+ ├── class_index.json class name -> id and color
246
+ ├── docs/ datasheet and figures
247
+ ├── real_ours/{train,validation,test}/data-*.parquet
248
+ ├── real_curated/train/data-*.parquet
249
+ ├── syn/train/data-000NN-of-00451.parquet
250
+ ├── syn/shard_index.json rows per class and environments in each syn shard
251
+ ├── synthetic_objects/<class>/<asset>/<asset>.{glb,ply,usdz}
252
+ ├── navable_<model>-<training data>.pth 28 checkpoints (see "Use the released checkpoints")
253
+ └── tools/ helpers and scripts
254
+ ```
255
+
256
+ `syn` rows are sorted by class, then environment, asset and trajectory. That is why one class lives in
257
+ a contiguous run of shards, listed in `syn/shard_index.json`.
258
+
259
+ ### Classes
260
+
261
+ Pixel values in every mask follow this table (also in `class_index.json`). The row counts are the
262
+ number of images whose primary class (`object_class`) is that class. Masks also label other classes
263
+ visible in the image.
264
+
265
+ | ID | Class | Color (RGB) | `syn` | `real_ours` train / val / test | `real_curated` | 3D assets |
266
+ |---:|---|---|---:|---:|---:|---:|
267
+ | 0 | background | (0, 0, 0) | | | | |
268
+ | 1 | `aps_button` | (220, 20, 60) | 62,690 | 206 / 23 / 66 | 0 | 15 |
269
+ | 2 | `bus_stop` | (255, 140, 0) | 60,770 | 205 / 23 / 62 | 0 | 61 |
270
+ | 3 | `bus_stop_sign` | (255, 215, 0) | 60,450 | 140 / 16 / 54 | 0 | – |
271
+ | 4 | `crosswalk` | (50, 205, 50) | 53,970 | 9 / 1 / 3 | 27,786 | – |
272
+ | 5 | `door_button` | (0, 191, 255) | 45,360 | 1,327 / 148 / 622 | 0 | 91 |
273
+ | 6 | `elevator` | (138, 43, 226) | 23,350 | 1,065 / 119 / 479 | 15 | 111 |
274
+ | 7 | `elevator_button` | (255, 105, 180) | 23,350 | 378 / 23 / 86 | 4,401 | 74 |
275
+ | 8 | `escalator` | (0, 128, 128) | 6,420 | 135 / 15 / 40 | 1,296 | 5 |
276
+ | 9 | `handrail` | (165, 42, 42) | 44,840 | 21 / 3 / 8 | 1,197 | 29 |
277
+ | 10 | `pedestrian_signal` | (75, 0, 130) | 44,650 | 217 / 25 / 62 | 6,650 | 66 |
278
+ | 11 | `revolving_door` | (255, 20, 147) | 25,010 | – | 0 | 48 |
279
+
280
+ `real_curated` has no `object_class` column; its counts are by the classes present in each mask.
281
+
282
+ ### Columns
283
+
284
+ <details>
285
+ <summary><b><code>real_ours</code></b>: collected real-world photos</summary>
286
+
287
+ | Column | Type | Content |
288
+ |---|---|---|
289
+ | `image` | image | RGB photo |
290
+ | `mask` | image | palettized PNG; pixel value = class id |
291
+ | `source` | string | `real_world` |
292
+ | `object_class` | string | the class the image was collected for |
293
+ | `class_id` | int32 | id of `object_class` |
294
+ | `location` | string | anonymized capture location (`location_00000083`); used to keep splits location-disjoint |
295
+ | `width`, `height` | int32 | image size |
296
+
297
+ </details>
298
+
299
+ <details>
300
+ <summary><b><code>real_curated</code></b>: curated public images</summary>
301
+
302
+ | Column | Type | Content |
303
+ |---|---|---|
304
+ | `image` | image | RGB image |
305
+ | `mask` | image | palettized PNG; pixel value = class id |
306
+ | `source` | string | `real_curated` |
307
+ | `source_dataset` | string | `curation` or `mapillary` |
308
+ | `split_origin` | string | the image's split in its source dataset |
309
+ | `width`, `height` | int32 | image size |
310
+
311
+ </details>
312
+
313
+ <details>
314
+ <summary><b><code>syn</code></b>: Isaac Sim renders</summary>
315
+
316
+ | Column | Type | Content |
317
+ |---|---|---|
318
+ | `image` | image | RGBA render, 1280×720 |
319
+ | `mask` | image | palettized PNG; pixel value = class id |
320
+ | `normals` | image | world-space surface normals (Z up), RGBA PNG as written by Isaac Sim Replicator (components mapped from [-1, 1] to [0, 255]) |
321
+ | `depth` | binary | distance to the image plane, float16 `.npy` bytes (720×1280), metres; `inf` where nothing was hit |
322
+ | `source` | string | `synthetic` |
323
+ | `object_class`, `class_id` | string, int32 | the class of the placed asset |
324
+ | `environment`, `sublocation` | string | scene (e.g. `downtown_west_sunset`) and the spot within it |
325
+ | `asset` | string | the placed 3D asset |
326
+ | `trajectory`, `frame_index` | string, int32 | camera trajectory and frame number |
327
+ | `bbox` | list of [x_min, y_min, x_max, y_max] | tight 2D boxes in pixels |
328
+ | `bbox_class_ids`, `occlusion_ratio`, `bbox_prim_paths` | lists | class id, occluded fraction and USD prim path of each 2D box |
329
+ | `bbox_3d` | list of [x_min, y_min, z_min, x_max, y_max, z_max] | 3D box extents in each object's local frame |
330
+ | `bbox_3d_transform` | list of 16 floats | row-major 4×4 local-to-world transform; translation in the last row, world units are centimetres |
331
+ | `bbox_3d_class_ids`, `bbox_3d_occlusion_ratio`, `bbox_3d_prim_paths` | lists | class id, occluded fraction and USD prim path of each 3D box |
332
+ | `width`, `height` | int32 | 1280, 720 |
333
+
334
+ </details>
335
+
336
+ **Decoding the synthetic modalities**
337
 
338
  ```python
339
+ import io, numpy as np
340
+
341
+ depth = np.load(io.BytesIO(row["depth"])) # float16 (720, 1280), metres (or nv.decode_depth)
342
+ normals = np.asarray(row["normals"])[..., :3] / 127.5 - 1.0 # approximate unit normals, world frame
343
+
344
+ # 3D box corners in world coordinates (centimetres)
345
+ ext = np.array(row["bbox_3d"][0]); T = np.array(row["bbox_3d_transform"][0]).reshape(4, 4)
346
+ corners = np.array([[x, y, z, 1] for x in ext[[0, 3]] for y in ext[[1, 4]] for z in ext[[2, 5]]])
347
+ corners_world = (corners @ T)[:, :3]
 
 
 
 
 
 
 
 
 
 
 
 
 
348
  ```
349
 
350
+ Camera intrinsics and poses are not included, so the 3D boxes can't be projected into the image.
351
 
352
+ ## How to use
 
 
 
 
 
 
353
 
354
+ ### Train a segmentation model (paper protocol)
355
 
356
+ The paper trains on `real_ours/train`, optionally adding `real_curated` (**+Cur**) and/or `syn` (**+Syn**).
357
+ It picks the checkpoint with the best foreground mIoU on `real_ours/validation` and reports on `real_ours/test`.
358
+ Models predict 12 labels (background + 11 classes). Real-world scores cover background and the 10 classes that have real examples.
359
 
360
+ Most training code reads image folders, so export the splits once:
 
 
 
 
 
 
 
 
 
 
 
 
361
 
362
+ ```bash
363
+ python tools/export.py --config real_ours --split train --out data/navable
364
+ python tools/export.py --config real_ours --split validation --out data/navable
365
+ python tools/export.py --config real_ours --split test --out data/navable
366
+ python tools/export.py --config syn --classes escalator --max-rows 5000 --out data/navable_syn # a synthetic subset
367
+ ```
368
 
369
+ This writes `img_dir/<split>/*.png` and `ann_dir/<split>/*.png` (label value = class id), the
370
+ [mmsegmentation](https://github.com/open-mmlab/mmsegmentation) layout. A matching mmseg 1.x dataset class:
371
 
372
+ ```python
373
+ from mmseg.datasets import BaseSegDataset
374
+ from mmseg.registry import DATASETS
375
+
376
+ @DATASETS.register_module()
377
+ class NavAbleDataset(BaseSegDataset):
378
+ METAINFO = dict(
379
+ classes=("background", "aps_button", "bus_stop", "bus_stop_sign", "crosswalk", "door_button", "elevator",
380
+ "elevator_button", "escalator", "handrail", "pedestrian_signal", "revolving_door"),
381
+ palette=[[0, 0, 0], [220, 20, 60], [255, 140, 0], [255, 215, 0], [50, 205, 50], [0, 191, 255],
382
+ [138, 43, 226], [255, 105, 180], [0, 128, 128], [165, 42, 42], [75, 0, 130], [255, 20, 147]])
383
+
384
+ def __init__(self, **kwargs):
385
+ super().__init__(img_suffix=".png", seg_map_suffix=".png", reduce_zero_label=False, **kwargs)
386
+ ```
387
+
388
+ <details>
389
+ <summary>Training settings used in the paper (Appendix A)</summary>
390
+
391
+ All models share one augmentation pipeline: random resize 0.5–2.0×, category-balanced random crop
392
+ (`cat_max_ratio=0.75`), horizontal flip (p = 0.5) and photometric distortion. They use a linear warm-up
393
+ then polynomial learning-rate decay, with class weights of 0.1 for background and 1.0 for each other class.
394
+ Iteration budgets grow with the training set; they are listed per model and data mix in the paper.
395
+
396
+ | Model | Backbone | Crop | Optimizer (base lr) | Weight decay | Batch |
397
  |---|---|---|---|---|---|
398
+ | DeepLabv3+ | ResNet-101 | 512 | SGD 5e-3, momentum 0.9 | 5e-4 | 8 |
399
+ | EncNet | ResNet-101 | 512 | SGD 5e-3, momentum 0.9 | 5e-4 | 8 |
400
+ | SegFormer | MiT-B5 | 640 | AdamW 1e-5 (head 10×) | 0.01 | 8 |
401
+ | Mask2Former | Swin-L | 640 | AdamW (backbone 0.1×) | 0.05 | 4–8 |
402
+ | SAN | CLIP ViT-B/16 | 640 | AdamW 1e-4 (encoder 0.1×) | 1e-4 | 24 |
403
+
404
+ The full training code and configs are in the [GitHub repository](https://github.com/DARoSLab/NavAble).
405
+
406
+ </details>
407
+
408
+ For a plain PyTorch loop, `nv.NavAbleSegmentation` wraps a split as a `Dataset` of tensors:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
409
 
410
  ```python
411
+ from torch.utils.data import DataLoader
412
+ train = nv.NavAbleSegmentation(nv.load("real_ours", "train"), size=(360, 640))
413
+ loader = DataLoader(train, batch_size=8, shuffle=True, num_workers=4)
414
+ batch = next(iter(loader)) # {"image": (8, 3, 360, 640) float, "mask": (8, 360, 640) int64}
415
  ```
416
 
417
+ ### Evaluate on the real-world test set
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
418
 
419
+ Save one label PNG per test image, named like the image (`row["image"]["path"]`, e.g.
420
+ `0f4eab9c-frame_003200.png`), with dataset class ids. Then run:
421
+
422
+ ```bash
423
+ python tools/evaluate.py --pred-dir preds/ --out results.json
424
+ ```
425
+
426
+ It prints per-class IoU, precision and recall, and the means the paper reports. **mIoU, mPrecision and mRecall
427
+ average over background plus the 10 classes with real-world examples.** It also prints the means over
428
+ the 10 classes alone. In code, the same metric is `nv.SegmentationMetrics`.
429
+
430
+ > **Comparing with the paper.** The `real_ours` masks here were rasterized from the annotation polygons
431
+ > with `pycocotools`. The paper's masks also include the polygon edge pixels, which differ on about 0.2% of pixels.
432
+ > On these masks, the released checkpoints score about 0.7–0.9 mIoU below the paper. Examples: SegFormer (Real)
433
+ > 60.20 vs 61.06, DeepLabv3+ (Real) 62.99 vs 63.84, Mask2Former (Real) 67.93 vs 68.84. Compare methods on the same masks.
434
+
435
+ ### Use the released checkpoints
436
 
437
+ The repo root holds 28 checkpoints: seven models, each trained on four data mixes. The table gives their
438
+ paper mIoU on `real_ours/test`, with the best mix per model in bold.
 
 
439
 
440
+ | Model (backbone) | Real | Real+Cur | Real+Syn | Real+Cur+Syn | File name |
441
+ |---|---:|---:|---:|---:|---|
442
+ | Mask2Former (Swin-L) | 68.84 | 71.29 | **72.67** | 70.75 | `navable_mask2former-<mix>.pth` |
443
+ | SAN (CLIP ViT-B/16) | 62.07 | 62.52 | 66.96 | **68.12** | `navable_san-vit-b16_<mix>.pth` (mix written with `-`) |
444
+ | SegFormer (MiT-B5) | 61.06 | 62.16 | 65.54 | **65.96** | `navable_segformer-<mix>.pth` |
445
+ | DeepLabv3+ (ResNet-101) | **63.84** | 59.49 | 52.63 | 59.84 | `navable_deeplabv3plus-<mix>.pth` |
446
+ | EncNet (ResNet-101) | **60.56** | 52.77 | 57.50 | 56.50 | `navable_encnet-<mix>.pth` |
447
+ | SegMAN | – | – | – | – | `navable_segman-<mix>.pth` (not in the paper's tables) |
448
+ | SAM 3 (segmentation head only) | – | – | – | – | `navable_sam3-headonly-<mix>.pth` (not in the paper's tables) |
449
 
450
+ `<mix>` is `real`, `real+cur`, `real+syn` or `real+cur+syn`. The +Syn checkpoints were trained on the
451
+ earlier render of `syn` (452,704 frames, before the 2026-11 re-release).
452
+
453
+ ```bash
454
+ python tools/download.py --checkpoints segformer --train-data real real+syn # or --checkpoints all --dry-run
455
+ python tools/eval_checkpoint.py --checkpoint navable/navable_segformer-real.pth # needs mmsegmentation 1.x
456
+ ```
457
+
458
+ > **Label order.** The checkpoints were trained with a different class order from the dataset masks:
459
+ > `background, elevator, elevator_button, door_button, crosswalk, pedestrian_signal, aps_button, bus_stop, bus_stop_sign, handrail, escalator, revolving_door`.
460
+ > SAN uses the same order without background. Map predictions with `nv.checkpoint_to_dataset_labels(pred)`
461
+ > (`has_background=False` for SAN), or pass `--label-order checkpoint` to `evaluate.py`. `eval_checkpoint.py`
462
+ > does this for you. It runs DeepLabv3+, EncNet, SegFormer and Mask2Former; SAN, SegMAN and SAM 3 need
463
+ > their own code (see GitHub). The checkpoint metadata predates the rename of class 11 and still uses its old name.
464
+
465
+ ### Work with the synthetic data
466
+
467
+ - **One class or a sample, not 816 GB:** `nv.load_syn(["handrail"])` streams only that class's shards.
468
+ `python tools/download.py --syn handrail` saves them locally, and `--syn all --max-shards 5` gives a quick 10 GB sample.
469
+ - **Depth and normals:** the frames work for depth or normal estimation and multi-task learning.
470
+ Depth is float16 in metres (at most 0.05% relative error from the float32 render).
471
+ - **Detection:** `bbox` / `bbox_class_ids` give tight 2D boxes for every labelled object, and `occlusion_ratio` lets you
472
+ drop heavily occluded ones. `bbox_3d` and `bbox_3d_transform` give 3D boxes in the Isaac Sim world frame.
473
+ - **Sim-to-real:** `environment` names include lighting variants (`_sunset`, `_night`), and `asset` identifies the 3D asset,
474
+ so you can hold out environments, lighting or assets.
475
+
476
+ ### Use the 3D assets
477
+
478
+ `synthetic_objects/<class>/<asset>/` holds each asset as a `.glb` mesh, a `.ply` Gaussian splat and a
479
+ `.usdz` bundle that Isaac Sim, Omniverse or AR viewers can load directly. The assets were reconstructed from web images with SAM 3D Objects
480
+ (see [How the data was made](#how-the-data-was-made)).
481
+
482
+ ```bash
483
+ python tools/download.py --assets door_button elevator_button
484
+ ```
485
+
486
+ ## Results from the paper
487
+
488
+ **Segmentation on the real-world test set** (mIoU, %). Adding synthetic data helps the transformer and
489
+ mask-classification models most. Zero-shot SAN without NavAble training reaches only 14.94 mIoU.
490
+
491
+ | Model | Real | Real+Cur | Real+Syn | Real+Cur+Syn | Best vs. Real |
492
+ |---|---:|---:|---:|---:|---:|
493
+ | SAN | 62.07 | 62.52 | 66.96 | **68.12** | +6.05 |
494
+ | Mask2Former | 68.84 | 71.29 | **72.67** | 70.75 | +3.83 |
495
+ | SegFormer | 61.06 | 62.16 | 65.54 | **65.96** | +4.90 |
496
+ | DeepLabv3+ | **63.84** | 59.49 | 52.63 | 59.84 | −4.00 |
497
+ | EncNet | **60.56** | 52.77 | 57.50 | 56.50 | −3.06 |
498
+
499
+ <details>
500
+ <summary>Vision-language models, zero-shot (363 test images, 10 classes)</summary>
501
+
502
+ | Model | Box AP | AP50 | Point in mask | PCK@0.05 | PCK@0.10 |
503
+ |---|---:|---:|---:|---:|---:|
504
+ | Gemini Robotics-ER 1.6 | 33.8 | 48.9 | 69.0 | 68.7 | 85.5 |
505
+ | Gemini 3 Pro | 31.5 | 45.9 | 68.6 | 61.2 | 79.4 |
506
+ | Qwen3-VL-32B | 20.0 | 29.7 | 59.0 | 59.7 | 79.2 |
507
+ | MolmoPoint-8B | – | – | 74.2 | 55.0 | 73.6 |
508
+
509
+ </details>
510
+
511
+ **On assistive robots.** On egocentric footage from a cane-shaped mobility device and a quadruped
512
+ guide-dog robot, SegFormer trained with synthetic data beats real-only training. mIoU rises from 11.94 to 16.17 on the cane
513
+ and from 21.56 to 23.08 on the quadruped, with the largest gains on pedestrian signals, APS buttons and handrails.
514
+
515
+ ![Robot evaluation](docs/robot-evaluation.png)
516
+
517
+ ## How the data was made
518
+
519
+ **Real data.** Public segmentation datasets have noisy, inconsistent or missing labels for these objects. We curated
520
+ 36,466 images from them for consistent masks and remapped them to one taxonomy (`real_curated`). We then collected 5,581 new photos
521
+ indoors and outdoors and annotated them with COCO polygons (`real_ours`). Test images come only from locations
522
+ not used for training or validation.
523
+
524
+ ![Data curation](docs/data-curation.png)
525
+
526
+ **Synthetic data.** The generator turns web images into simulation-ready assets. A vision-language model
527
+ filters candidate images, and Grounding DINO and SAM isolate the object. SAM 3D Objects reconstructs a textured
528
+ 3D asset, which is placed in Isaac Sim scenes and rendered along configurable camera trajectories under day,
529
+ sunset and night lighting. Replicator writes RGB, class masks, depth, surface normals and 2D/3D boxes for every frame.
530
+
531
+ ![Synthetic data generation pipeline](docs/generation-pipeline.png)
532
+
533
+ ## Tools
534
+
535
+ | Script | What it does |
536
+ |---|---|
537
+ | [`tools/navable.py`](tools/navable.py) | Helpers: class names and palette, loaders (`load`, `load_syn`), depth / mask / box decoding and drawing, the paper's metrics, checkpoint label remapping, a PyTorch `Dataset`. |
538
+ | [`tools/download.py`](tools/download.py) | Download only what you need (real data, `syn` shards by class, 3D assets, checkpoints); prints sizes first. |
539
+ | [`tools/explore.py`](tools/explore.py) | Save a grid of random samples with masks and, for `syn`, depth, normals and boxes. |
540
+ | [`tools/export.py`](tools/export.py) | Write a config/split as image + label folders (mmsegmentation layout), optionally with depth and normals. |
541
+ | [`tools/evaluate.py`](tools/evaluate.py) | Score prediction PNGs on `real_ours` with the paper's metrics. |
542
+ | [`tools/eval_checkpoint.py`](tools/eval_checkpoint.py) | Run a released checkpoint on `real_ours/test`. |
543
+
544
+ They need only `datasets`, `huggingface_hub`, `pyarrow`, `numpy` and `Pillow`; PyTorch and mmsegmentation are needed only for training and checkpoint evaluation.
545
+ Training code for the paper's models is in the [GitHub repository](https://github.com/DARoSLab/NavAble).
546
+
547
+ ## Dataset card
548
+
549
+ **Intended uses.** Training and evaluating perception for assistive navigation, such as segmentation and detection of
550
+ accessibility-critical infrastructure; sim-to-real transfer; depth and normal estimation on synthetic data; and
551
+ asset-based synthetic data generation.
552
+
553
+ **Out-of-scope uses.** Surveillance or identifying people. Models trained on NavAble are not validated for unsupervised
554
+ safety-critical deployment; BLV navigation aids need further testing with their users.
555
+
556
+ **Personal and sensitive information.** The synthetic data contains no real people. Real photos were taken in public
557
+ spaces for accessibility research; capture locations are anonymized ids.
558
+
559
+ **Known limitations.**
560
+ - Real images are 640×360 while synthetic frames are 1280×720; resize to a common input size.
561
+ - Some classes have few real examples (for example `crosswalk` and `handrail` in `real_ours`); report per-class scores.
562
+ - `revolving_door` is synthetic-only, so real-world metrics use the other 10 classes.
563
+ - Synthetic textures and lighting do not match the real world exactly.
564
+
565
+ **License.** [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).
566
+
567
+ **More.** [Datasheet for Datasets](docs/datasheet.md) · [Class index](class_index.json) · Croissant metadata (the *Croissant* button on the dataset page).
568
+ Questions and issues: open a discussion on this page or an issue on [GitHub](https://github.com/DARoSLab/NavAble).
569
 
570
  ## Citation
571
 
 
581
  url = {https://neurips.cc/virtual/2026/loc/sydney/poster/138983}
582
  }
583
  ```
584
+
585
+ ## Changelog
586
+
587
+ - **2026-11**: `syn` re-rendered with surface normals, depth and 3D bounding boxes (450,860 frames). Class 11 renamed to
588
+ `revolving_door` (same id and color; 3D assets moved to `synthetic_objects/revolving_door/`). Added `tools/`,
589
+ `syn/shard_index.json` and this card. Other configs are unchanged.
590
+ - **2026-05**: first release.
docs/{hero.png → data-curation.png} RENAMED
File without changes
docs/datasheet.md CHANGED
@@ -1,4 +1,7 @@
1
- # Datasheet: BLV Object Recognition (Synthetic + Real-World)
 
 
 
2
 
3
  Following the structure of *Datasheets for Datasets* (Gebru et al., 2018).
4
 
@@ -19,10 +22,11 @@ DARoS Lab.
19
 
20
  **What do instances represent?**
21
  Each instance is a single image with a paired pixel-level segmentation mask.
22
- Synthetic instances additionally carry 2D bounding boxes per object.
 
23
 
24
  **How many instances are there?**
25
- - `syn/train`: 452704
26
  - `real_ours/train`: 3703
27
  - `real_ours/validation`: 396
28
  - `real_ours/test`: 1482
@@ -46,9 +50,12 @@ release.
46
 
47
  ## Collection process
48
 
49
- **Synthetic.** Generated in NVIDIA IsaacSim using Replicator. Each trajectory
50
- samples an asset, environment, and lighting condition, and records
51
- RGB + semantic mask + 2D bounding box per frame.
 
 
 
52
 
53
  **Real-world.** Photographs captured by data collectors at distinct physical
54
  locations covering the 10 shared object classes; annotated with COCO-format
@@ -56,20 +63,24 @@ polygon segmentations.
56
 
57
  ## Preprocessing / cleaning / labeling
58
 
59
- See `README.md` and `scripts/build_hf_dataset.py`. The on-disk layout
60
- re-encodes synthetic RGBA-coded masks into a single global palettized format
61
- and rasterizes real-world COCO polygons into the same format. Source RGB PNGs
62
- are not re-encoded.
63
 
64
  ## Uses
65
 
66
  **Has the dataset been used for any tasks already?**
67
- The dataset will accompany a paper at the NeurIPS 2026 Datasets & Benchmarks
68
- track (submission pending).
 
 
 
69
 
70
  **What other tasks could the dataset be used for?**
71
  Sim-to-real transfer studies, robustness analysis under lighting conditions,
72
- multi-task learning combining detection and segmentation.
 
73
 
74
  **Are there tasks for which the dataset should not be used?**
75
  The dataset must not be used for surveillance or identification of
@@ -80,4 +91,12 @@ accessibility research.
80
  ## Distribution
81
 
82
  The dataset is hosted on Hugging Face at `NavAble/NeurIPS_2026_BLV` and
83
- licensed under CC BY 4.0.
 
 
 
 
 
 
 
 
 
1
+ # Datasheet: NavAble
2
+
3
+ NavAble: A Large-Scale Dataset and Synthetic Data Generation Pipeline for Blind Navigation
4
+ (NeurIPS 2026; [arXiv:2610.19330](https://arxiv.org/abs/2610.19330); [project page](https://guidedogrobot-navable.github.io/)).
5
 
6
  Following the structure of *Datasheets for Datasets* (Gebru et al., 2018).
7
 
 
22
 
23
  **What do instances represent?**
24
  Each instance is a single image with a paired pixel-level segmentation mask.
25
+ Synthetic instances additionally carry a depth map, a surface-normal map, and
26
+ 2D and 3D bounding boxes (with occlusion ratios and USD prim paths) per object.
27
 
28
  **How many instances are there?**
29
+ - `syn/train`: 450860
30
  - `real_ours/train`: 3703
31
  - `real_ours/validation`: 396
32
  - `real_ours/test`: 1482
 
50
 
51
  ## Collection process
52
 
53
+ **Synthetic.** Web images of each class are filtered with a vision-language
54
+ model, isolated with Grounding DINO and SAM, and reconstructed into textured 3D
55
+ assets with SAM 3D Objects. The assets are placed in NVIDIA Isaac Sim scenes (37
56
+ environments with day, sunset and night variants) and rendered with Replicator
57
+ along camera trajectories. Each frame records RGB, a class mask, depth, surface
58
+ normals, and 2D and 3D bounding boxes.
59
 
60
  **Real-world.** Photographs captured by data collectors at distinct physical
61
  locations covering the 10 shared object classes; annotated with COCO-format
 
63
 
64
  ## Preprocessing / cleaning / labeling
65
 
66
+ See `README.md`. Synthetic RGBA-coded masks are re-encoded into one global
67
+ palettized format, and real-world COCO polygons are rasterized into the same
68
+ format. RGB and surface-normal PNGs are stored as rendered or captured. Depth is
69
+ converted from float32 to float16 (at most 0.05% relative error).
70
 
71
  ## Uses
72
 
73
  **Has the dataset been used for any tasks already?**
74
+ Semantic segmentation of accessibility-critical objects, reported in the
75
+ NeurIPS 2026 paper above (five segmentation models under four training-data
76
+ mixes, zero-shot vision-language models, and transfer to a cane-shaped device
77
+ and a quadruped guide-dog robot). The trained checkpoints are released with
78
+ the dataset.
79
 
80
  **What other tasks could the dataset be used for?**
81
  Sim-to-real transfer studies, robustness analysis under lighting conditions,
82
+ object detection, depth and surface-normal estimation, multi-task learning, and
83
+ synthetic data generation with the released 3D assets.
84
 
85
  **Are there tasks for which the dataset should not be used?**
86
  The dataset must not be used for surveillance or identification of
 
91
  ## Distribution
92
 
93
  The dataset is hosted on Hugging Face at `NavAble/NeurIPS_2026_BLV` and
94
+ licensed under CC BY 4.0. Code related to the paper is at
95
+ https://github.com/DARoSLab/NavAble.
96
+
97
+ ## Maintenance
98
+
99
+ The dataset is maintained by the authors (DARoS Lab, University of
100
+ Massachusetts Amherst). Questions and corrections: open a discussion on the
101
+ Hugging Face dataset page or an issue on GitHub. Changes are recorded in the
102
+ changelog of `README.md`.
docs/generation-pipeline.png ADDED

Git LFS Details

  • SHA256: 98f313ff3c40c6c65ecb9bea526b8df840429949208df50644553cf4e01d5d46
  • Pointer size: 131 Bytes
  • Size of remote file: 436 kB
docs/robot-evaluation.png ADDED

Git LFS Details

  • SHA256: 7add9d19055132660a87aefc658fc8ba738278239ebb5b864d061cb979e70e85
  • Pointer size: 131 Bytes
  • Size of remote file: 923 kB
docs/teaser.png ADDED

Git LFS Details

  • SHA256: 32b6b742033caf02bd3c162e36c7b2b3e43061e35d5a17d4d7b2f5d89f3fa9ed
  • Pointer size: 131 Bytes
  • Size of remote file: 852 kB
syn/shard_index.json ADDED
@@ -0,0 +1,5148 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "description": "Per-shard contents of syn/train, so a subset (e.g. one class) can be downloaded without the full 816 GB. Rows are grouped by class, then environment, asset and trajectory.",
3
+ "num_shards": 451,
4
+ "num_rows": 450860,
5
+ "shards": [
6
+ {
7
+ "file": "syn/train/data-00000-of-00451.parquet",
8
+ "rows": 1000,
9
+ "bytes": 2420838308,
10
+ "classes": {
11
+ "aps_button": 1000
12
+ },
13
+ "environments": [
14
+ "downtown_west"
15
+ ]
16
+ },
17
+ {
18
+ "file": "syn/train/data-00001-of-00451.parquet",
19
+ "rows": 1000,
20
+ "bytes": 2140517627,
21
+ "classes": {
22
+ "aps_button": 1000
23
+ },
24
+ "environments": [
25
+ "downtown_west"
26
+ ]
27
+ },
28
+ {
29
+ "file": "syn/train/data-00002-of-00451.parquet",
30
+ "rows": 1000,
31
+ "bytes": 2248254366,
32
+ "classes": {
33
+ "aps_button": 1000
34
+ },
35
+ "environments": [
36
+ "downtown_west",
37
+ "downtown_west_night"
38
+ ]
39
+ },
40
+ {
41
+ "file": "syn/train/data-00003-of-00451.parquet",
42
+ "rows": 1000,
43
+ "bytes": 1977347815,
44
+ "classes": {
45
+ "aps_button": 1000
46
+ },
47
+ "environments": [
48
+ "downtown_west_night"
49
+ ]
50
+ },
51
+ {
52
+ "file": "syn/train/data-00004-of-00451.parquet",
53
+ "rows": 1000,
54
+ "bytes": 1782252831,
55
+ "classes": {
56
+ "aps_button": 1000
57
+ },
58
+ "environments": [
59
+ "downtown_west_night"
60
+ ]
61
+ },
62
+ {
63
+ "file": "syn/train/data-00005-of-00451.parquet",
64
+ "rows": 1000,
65
+ "bytes": 2070663388,
66
+ "classes": {
67
+ "aps_button": 1000
68
+ },
69
+ "environments": [
70
+ "downtown_west_night",
71
+ "downtown_west_sunset"
72
+ ]
73
+ },
74
+ {
75
+ "file": "syn/train/data-00006-of-00451.parquet",
76
+ "rows": 1000,
77
+ "bytes": 2215434433,
78
+ "classes": {
79
+ "aps_button": 1000
80
+ },
81
+ "environments": [
82
+ "downtown_west_sunset"
83
+ ]
84
+ },
85
+ {
86
+ "file": "syn/train/data-00007-of-00451.parquet",
87
+ "rows": 1000,
88
+ "bytes": 2206115630,
89
+ "classes": {
90
+ "aps_button": 1000
91
+ },
92
+ "environments": [
93
+ "downtown_west_sunset"
94
+ ]
95
+ },
96
+ {
97
+ "file": "syn/train/data-00008-of-00451.parquet",
98
+ "rows": 1000,
99
+ "bytes": 1888827337,
100
+ "classes": {
101
+ "aps_button": 1000
102
+ },
103
+ "environments": [
104
+ "downtown_west_sunset",
105
+ "hong_kong_street"
106
+ ]
107
+ },
108
+ {
109
+ "file": "syn/train/data-00009-of-00451.parquet",
110
+ "rows": 1000,
111
+ "bytes": 1865418197,
112
+ "classes": {
113
+ "aps_button": 1000
114
+ },
115
+ "environments": [
116
+ "hong_kong_street"
117
+ ]
118
+ },
119
+ {
120
+ "file": "syn/train/data-00010-of-00451.parquet",
121
+ "rows": 1000,
122
+ "bytes": 1771664368,
123
+ "classes": {
124
+ "aps_button": 1000
125
+ },
126
+ "environments": [
127
+ "hong_kong_street",
128
+ "hong_kong_street_sunset"
129
+ ]
130
+ },
131
+ {
132
+ "file": "syn/train/data-00011-of-00451.parquet",
133
+ "rows": 1000,
134
+ "bytes": 1834960163,
135
+ "classes": {
136
+ "aps_button": 1000
137
+ },
138
+ "environments": [
139
+ "hong_kong_street_sunset"
140
+ ]
141
+ },
142
+ {
143
+ "file": "syn/train/data-00012-of-00451.parquet",
144
+ "rows": 1000,
145
+ "bytes": 2260023777,
146
+ "classes": {
147
+ "aps_button": 1000
148
+ },
149
+ "environments": [
150
+ "hong_kong_street_sunset",
151
+ "miami_beach"
152
+ ]
153
+ },
154
+ {
155
+ "file": "syn/train/data-00013-of-00451.parquet",
156
+ "rows": 1000,
157
+ "bytes": 2554491404,
158
+ "classes": {
159
+ "aps_button": 1000
160
+ },
161
+ "environments": [
162
+ "miami_beach"
163
+ ]
164
+ },
165
+ {
166
+ "file": "syn/train/data-00014-of-00451.parquet",
167
+ "rows": 1000,
168
+ "bytes": 2590811983,
169
+ "classes": {
170
+ "aps_button": 1000
171
+ },
172
+ "environments": [
173
+ "miami_beach"
174
+ ]
175
+ },
176
+ {
177
+ "file": "syn/train/data-00015-of-00451.parquet",
178
+ "rows": 1000,
179
+ "bytes": 2438625277,
180
+ "classes": {
181
+ "aps_button": 1000
182
+ },
183
+ "environments": [
184
+ "miami_beach"
185
+ ]
186
+ },
187
+ {
188
+ "file": "syn/train/data-00016-of-00451.parquet",
189
+ "rows": 1000,
190
+ "bytes": 2260001211,
191
+ "classes": {
192
+ "aps_button": 1000
193
+ },
194
+ "environments": [
195
+ "miami_beach_night"
196
+ ]
197
+ },
198
+ {
199
+ "file": "syn/train/data-00017-of-00451.parquet",
200
+ "rows": 1000,
201
+ "bytes": 2570696488,
202
+ "classes": {
203
+ "aps_button": 1000
204
+ },
205
+ "environments": [
206
+ "miami_beach_night"
207
+ ]
208
+ },
209
+ {
210
+ "file": "syn/train/data-00018-of-00451.parquet",
211
+ "rows": 1000,
212
+ "bytes": 2526972652,
213
+ "classes": {
214
+ "aps_button": 1000
215
+ },
216
+ "environments": [
217
+ "miami_beach_night"
218
+ ]
219
+ },
220
+ {
221
+ "file": "syn/train/data-00019-of-00451.parquet",
222
+ "rows": 1000,
223
+ "bytes": 2428406338,
224
+ "classes": {
225
+ "aps_button": 1000
226
+ },
227
+ "environments": [
228
+ "miami_beach_night",
229
+ "neighborhood"
230
+ ]
231
+ },
232
+ {
233
+ "file": "syn/train/data-00020-of-00451.parquet",
234
+ "rows": 1000,
235
+ "bytes": 2483193387,
236
+ "classes": {
237
+ "aps_button": 1000
238
+ },
239
+ "environments": [
240
+ "neighborhood",
241
+ "neighborhood_sunset"
242
+ ]
243
+ },
244
+ {
245
+ "file": "syn/train/data-00021-of-00451.parquet",
246
+ "rows": 1000,
247
+ "bytes": 2557115759,
248
+ "classes": {
249
+ "aps_button": 1000
250
+ },
251
+ "environments": [
252
+ "neighborhood_sunset"
253
+ ]
254
+ },
255
+ {
256
+ "file": "syn/train/data-00022-of-00451.parquet",
257
+ "rows": 1000,
258
+ "bytes": 2102583093,
259
+ "classes": {
260
+ "aps_button": 1000
261
+ },
262
+ "environments": [
263
+ "neighborhood_sunset",
264
+ "night_city"
265
+ ]
266
+ },
267
+ {
268
+ "file": "syn/train/data-00023-of-00451.parquet",
269
+ "rows": 1000,
270
+ "bytes": 1792302500,
271
+ "classes": {
272
+ "aps_button": 1000
273
+ },
274
+ "environments": [
275
+ "night_city"
276
+ ]
277
+ },
278
+ {
279
+ "file": "syn/train/data-00024-of-00451.parquet",
280
+ "rows": 1000,
281
+ "bytes": 1627650153,
282
+ "classes": {
283
+ "aps_button": 1000
284
+ },
285
+ "environments": [
286
+ "night_city",
287
+ "nordic_harbour"
288
+ ]
289
+ },
290
+ {
291
+ "file": "syn/train/data-00025-of-00451.parquet",
292
+ "rows": 1000,
293
+ "bytes": 1740990742,
294
+ "classes": {
295
+ "aps_button": 1000
296
+ },
297
+ "environments": [
298
+ "nordic_harbour"
299
+ ]
300
+ },
301
+ {
302
+ "file": "syn/train/data-00026-of-00451.parquet",
303
+ "rows": 1000,
304
+ "bytes": 1991840433,
305
+ "classes": {
306
+ "aps_button": 1000
307
+ },
308
+ "environments": [
309
+ "nordic_harbour"
310
+ ]
311
+ },
312
+ {
313
+ "file": "syn/train/data-00027-of-00451.parquet",
314
+ "rows": 1000,
315
+ "bytes": 2424745876,
316
+ "classes": {
317
+ "aps_button": 1000
318
+ },
319
+ "environments": [
320
+ "nordic_harbour",
321
+ "nordic_harbour_night"
322
+ ]
323
+ },
324
+ {
325
+ "file": "syn/train/data-00028-of-00451.parquet",
326
+ "rows": 1000,
327
+ "bytes": 1192658390,
328
+ "classes": {
329
+ "aps_button": 1000
330
+ },
331
+ "environments": [
332
+ "nordic_harbour_night"
333
+ ]
334
+ },
335
+ {
336
+ "file": "syn/train/data-00029-of-00451.parquet",
337
+ "rows": 1000,
338
+ "bytes": 1297173459,
339
+ "classes": {
340
+ "aps_button": 1000
341
+ },
342
+ "environments": [
343
+ "nordic_harbour_night"
344
+ ]
345
+ },
346
+ {
347
+ "file": "syn/train/data-00030-of-00451.parquet",
348
+ "rows": 1000,
349
+ "bytes": 1838589408,
350
+ "classes": {
351
+ "aps_button": 1000
352
+ },
353
+ "environments": [
354
+ "nordic_harbour_night"
355
+ ]
356
+ },
357
+ {
358
+ "file": "syn/train/data-00031-of-00451.parquet",
359
+ "rows": 1000,
360
+ "bytes": 1633494989,
361
+ "classes": {
362
+ "aps_button": 1000
363
+ },
364
+ "environments": [
365
+ "nordic_harbour_night",
366
+ "nordic_harbour_sunset"
367
+ ]
368
+ },
369
+ {
370
+ "file": "syn/train/data-00032-of-00451.parquet",
371
+ "rows": 1000,
372
+ "bytes": 1722345088,
373
+ "classes": {
374
+ "aps_button": 1000
375
+ },
376
+ "environments": [
377
+ "nordic_harbour_sunset"
378
+ ]
379
+ },
380
+ {
381
+ "file": "syn/train/data-00033-of-00451.parquet",
382
+ "rows": 1000,
383
+ "bytes": 2186765528,
384
+ "classes": {
385
+ "aps_button": 1000
386
+ },
387
+ "environments": [
388
+ "nordic_harbour_sunset"
389
+ ]
390
+ },
391
+ {
392
+ "file": "syn/train/data-00034-of-00451.parquet",
393
+ "rows": 1000,
394
+ "bytes": 2280679646,
395
+ "classes": {
396
+ "aps_button": 1000
397
+ },
398
+ "environments": [
399
+ "nordic_harbour_sunset",
400
+ "stylized_city"
401
+ ]
402
+ },
403
+ {
404
+ "file": "syn/train/data-00035-of-00451.parquet",
405
+ "rows": 1000,
406
+ "bytes": 1884676830,
407
+ "classes": {
408
+ "aps_button": 1000
409
+ },
410
+ "environments": [
411
+ "stylized_city"
412
+ ]
413
+ },
414
+ {
415
+ "file": "syn/train/data-00036-of-00451.parquet",
416
+ "rows": 1000,
417
+ "bytes": 1838276262,
418
+ "classes": {
419
+ "aps_button": 1000
420
+ },
421
+ "environments": [
422
+ "stylized_city"
423
+ ]
424
+ },
425
+ {
426
+ "file": "syn/train/data-00037-of-00451.parquet",
427
+ "rows": 1000,
428
+ "bytes": 1972168214,
429
+ "classes": {
430
+ "aps_button": 1000
431
+ },
432
+ "environments": [
433
+ "stylized_city"
434
+ ]
435
+ },
436
+ {
437
+ "file": "syn/train/data-00038-of-00451.parquet",
438
+ "rows": 1000,
439
+ "bytes": 1734743513,
440
+ "classes": {
441
+ "aps_button": 1000
442
+ },
443
+ "environments": [
444
+ "stylized_city",
445
+ "stylized_city_night"
446
+ ]
447
+ },
448
+ {
449
+ "file": "syn/train/data-00039-of-00451.parquet",
450
+ "rows": 1000,
451
+ "bytes": 1728899069,
452
+ "classes": {
453
+ "aps_button": 1000
454
+ },
455
+ "environments": [
456
+ "stylized_city_night"
457
+ ]
458
+ },
459
+ {
460
+ "file": "syn/train/data-00040-of-00451.parquet",
461
+ "rows": 1000,
462
+ "bytes": 1555986322,
463
+ "classes": {
464
+ "aps_button": 1000
465
+ },
466
+ "environments": [
467
+ "stylized_city_night"
468
+ ]
469
+ },
470
+ {
471
+ "file": "syn/train/data-00041-of-00451.parquet",
472
+ "rows": 1000,
473
+ "bytes": 1656021107,
474
+ "classes": {
475
+ "aps_button": 1000
476
+ },
477
+ "environments": [
478
+ "stylized_city_night"
479
+ ]
480
+ },
481
+ {
482
+ "file": "syn/train/data-00042-of-00451.parquet",
483
+ "rows": 1000,
484
+ "bytes": 1560564885,
485
+ "classes": {
486
+ "aps_button": 1000
487
+ },
488
+ "environments": [
489
+ "stylized_city_night",
490
+ "stylized_city_sunset"
491
+ ]
492
+ },
493
+ {
494
+ "file": "syn/train/data-00043-of-00451.parquet",
495
+ "rows": 1000,
496
+ "bytes": 2078413922,
497
+ "classes": {
498
+ "aps_button": 1000
499
+ },
500
+ "environments": [
501
+ "stylized_city_sunset"
502
+ ]
503
+ },
504
+ {
505
+ "file": "syn/train/data-00044-of-00451.parquet",
506
+ "rows": 1000,
507
+ "bytes": 1723094561,
508
+ "classes": {
509
+ "aps_button": 1000
510
+ },
511
+ "environments": [
512
+ "stylized_city_sunset"
513
+ ]
514
+ },
515
+ {
516
+ "file": "syn/train/data-00045-of-00451.parquet",
517
+ "rows": 1000,
518
+ "bytes": 1904204123,
519
+ "classes": {
520
+ "aps_button": 1000
521
+ },
522
+ "environments": [
523
+ "stylized_city_sunset"
524
+ ]
525
+ },
526
+ {
527
+ "file": "syn/train/data-00046-of-00451.parquet",
528
+ "rows": 1000,
529
+ "bytes": 1792834649,
530
+ "classes": {
531
+ "aps_button": 1000
532
+ },
533
+ "environments": [
534
+ "stylized_city_sunset"
535
+ ]
536
+ },
537
+ {
538
+ "file": "syn/train/data-00047-of-00451.parquet",
539
+ "rows": 1000,
540
+ "bytes": 2379918426,
541
+ "classes": {
542
+ "aps_button": 1000
543
+ },
544
+ "environments": [
545
+ "stylized_city_sunset",
546
+ "suburbs"
547
+ ]
548
+ },
549
+ {
550
+ "file": "syn/train/data-00048-of-00451.parquet",
551
+ "rows": 1000,
552
+ "bytes": 2330576644,
553
+ "classes": {
554
+ "aps_button": 1000
555
+ },
556
+ "environments": [
557
+ "suburbs"
558
+ ]
559
+ },
560
+ {
561
+ "file": "syn/train/data-00049-of-00451.parquet",
562
+ "rows": 1000,
563
+ "bytes": 2249902713,
564
+ "classes": {
565
+ "aps_button": 1000
566
+ },
567
+ "environments": [
568
+ "suburbs"
569
+ ]
570
+ },
571
+ {
572
+ "file": "syn/train/data-00050-of-00451.parquet",
573
+ "rows": 1000,
574
+ "bytes": 2276691781,
575
+ "classes": {
576
+ "aps_button": 1000
577
+ },
578
+ "environments": [
579
+ "suburbs",
580
+ "suburbs_night"
581
+ ]
582
+ },
583
+ {
584
+ "file": "syn/train/data-00051-of-00451.parquet",
585
+ "rows": 1000,
586
+ "bytes": 1909333235,
587
+ "classes": {
588
+ "aps_button": 1000
589
+ },
590
+ "environments": [
591
+ "suburbs_night"
592
+ ]
593
+ },
594
+ {
595
+ "file": "syn/train/data-00052-of-00451.parquet",
596
+ "rows": 1000,
597
+ "bytes": 1772386472,
598
+ "classes": {
599
+ "aps_button": 1000
600
+ },
601
+ "environments": [
602
+ "suburbs_night"
603
+ ]
604
+ },
605
+ {
606
+ "file": "syn/train/data-00053-of-00451.parquet",
607
+ "rows": 1000,
608
+ "bytes": 2097192080,
609
+ "classes": {
610
+ "aps_button": 1000
611
+ },
612
+ "environments": [
613
+ "suburbs_night",
614
+ "suburbs_sunset"
615
+ ]
616
+ },
617
+ {
618
+ "file": "syn/train/data-00054-of-00451.parquet",
619
+ "rows": 1000,
620
+ "bytes": 2359805534,
621
+ "classes": {
622
+ "aps_button": 1000
623
+ },
624
+ "environments": [
625
+ "suburbs_sunset"
626
+ ]
627
+ },
628
+ {
629
+ "file": "syn/train/data-00055-of-00451.parquet",
630
+ "rows": 1000,
631
+ "bytes": 2256633389,
632
+ "classes": {
633
+ "aps_button": 1000
634
+ },
635
+ "environments": [
636
+ "suburbs_sunset"
637
+ ]
638
+ },
639
+ {
640
+ "file": "syn/train/data-00056-of-00451.parquet",
641
+ "rows": 1000,
642
+ "bytes": 2125127548,
643
+ "classes": {
644
+ "aps_button": 1000
645
+ },
646
+ "environments": [
647
+ "suburbs_sunset"
648
+ ]
649
+ },
650
+ {
651
+ "file": "syn/train/data-00057-of-00451.parquet",
652
+ "rows": 1000,
653
+ "bytes": 1953401987,
654
+ "classes": {
655
+ "aps_button": 1000
656
+ },
657
+ "environments": [
658
+ "suburbs_sunset",
659
+ "tokyo_street"
660
+ ]
661
+ },
662
+ {
663
+ "file": "syn/train/data-00058-of-00451.parquet",
664
+ "rows": 1000,
665
+ "bytes": 1955684055,
666
+ "classes": {
667
+ "aps_button": 1000
668
+ },
669
+ "environments": [
670
+ "tokyo_street",
671
+ "tokyo_street_sunset"
672
+ ]
673
+ },
674
+ {
675
+ "file": "syn/train/data-00059-of-00451.parquet",
676
+ "rows": 1000,
677
+ "bytes": 2066626092,
678
+ "classes": {
679
+ "aps_button": 1000
680
+ },
681
+ "environments": [
682
+ "tokyo_street_sunset",
683
+ "village"
684
+ ]
685
+ },
686
+ {
687
+ "file": "syn/train/data-00060-of-00451.parquet",
688
+ "rows": 1000,
689
+ "bytes": 2283070685,
690
+ "classes": {
691
+ "aps_button": 1000
692
+ },
693
+ "environments": [
694
+ "village"
695
+ ]
696
+ },
697
+ {
698
+ "file": "syn/train/data-00061-of-00451.parquet",
699
+ "rows": 1000,
700
+ "bytes": 2205772364,
701
+ "classes": {
702
+ "aps_button": 1000
703
+ },
704
+ "environments": [
705
+ "village_sunset"
706
+ ]
707
+ },
708
+ {
709
+ "file": "syn/train/data-00062-of-00451.parquet",
710
+ "rows": 1000,
711
+ "bytes": 2192178288,
712
+ "classes": {
713
+ "aps_button": 690,
714
+ "bus_stop": 310
715
+ },
716
+ "environments": [
717
+ "downtown_west",
718
+ "village_sunset"
719
+ ]
720
+ },
721
+ {
722
+ "file": "syn/train/data-00063-of-00451.parquet",
723
+ "rows": 1000,
724
+ "bytes": 1855207853,
725
+ "classes": {
726
+ "bus_stop": 1000
727
+ },
728
+ "environments": [
729
+ "downtown_west"
730
+ ]
731
+ },
732
+ {
733
+ "file": "syn/train/data-00064-of-00451.parquet",
734
+ "rows": 1000,
735
+ "bytes": 2018432461,
736
+ "classes": {
737
+ "bus_stop": 1000
738
+ },
739
+ "environments": [
740
+ "downtown_west"
741
+ ]
742
+ },
743
+ {
744
+ "file": "syn/train/data-00065-of-00451.parquet",
745
+ "rows": 1000,
746
+ "bytes": 1990746693,
747
+ "classes": {
748
+ "bus_stop": 1000
749
+ },
750
+ "environments": [
751
+ "downtown_west"
752
+ ]
753
+ },
754
+ {
755
+ "file": "syn/train/data-00066-of-00451.parquet",
756
+ "rows": 1000,
757
+ "bytes": 2565598207,
758
+ "classes": {
759
+ "bus_stop": 1000
760
+ },
761
+ "environments": [
762
+ "downtown_west"
763
+ ]
764
+ },
765
+ {
766
+ "file": "syn/train/data-00067-of-00451.parquet",
767
+ "rows": 1000,
768
+ "bytes": 1838879632,
769
+ "classes": {
770
+ "bus_stop": 1000
771
+ },
772
+ "environments": [
773
+ "downtown_west",
774
+ "downtown_west_night"
775
+ ]
776
+ },
777
+ {
778
+ "file": "syn/train/data-00068-of-00451.parquet",
779
+ "rows": 1000,
780
+ "bytes": 1543033855,
781
+ "classes": {
782
+ "bus_stop": 1000
783
+ },
784
+ "environments": [
785
+ "downtown_west_night"
786
+ ]
787
+ },
788
+ {
789
+ "file": "syn/train/data-00069-of-00451.parquet",
790
+ "rows": 1000,
791
+ "bytes": 1566484245,
792
+ "classes": {
793
+ "bus_stop": 1000
794
+ },
795
+ "environments": [
796
+ "downtown_west_night"
797
+ ]
798
+ },
799
+ {
800
+ "file": "syn/train/data-00070-of-00451.parquet",
801
+ "rows": 1000,
802
+ "bytes": 1781869135,
803
+ "classes": {
804
+ "bus_stop": 1000
805
+ },
806
+ "environments": [
807
+ "downtown_west_night"
808
+ ]
809
+ },
810
+ {
811
+ "file": "syn/train/data-00071-of-00451.parquet",
812
+ "rows": 1000,
813
+ "bytes": 1978215003,
814
+ "classes": {
815
+ "bus_stop": 1000
816
+ },
817
+ "environments": [
818
+ "downtown_west_night"
819
+ ]
820
+ },
821
+ {
822
+ "file": "syn/train/data-00072-of-00451.parquet",
823
+ "rows": 1000,
824
+ "bytes": 1831793022,
825
+ "classes": {
826
+ "bus_stop": 1000
827
+ },
828
+ "environments": [
829
+ "downtown_west_night",
830
+ "downtown_west_sunset"
831
+ ]
832
+ },
833
+ {
834
+ "file": "syn/train/data-00073-of-00451.parquet",
835
+ "rows": 1000,
836
+ "bytes": 1901720935,
837
+ "classes": {
838
+ "bus_stop": 1000
839
+ },
840
+ "environments": [
841
+ "downtown_west_sunset"
842
+ ]
843
+ },
844
+ {
845
+ "file": "syn/train/data-00074-of-00451.parquet",
846
+ "rows": 1000,
847
+ "bytes": 1714641193,
848
+ "classes": {
849
+ "bus_stop": 1000
850
+ },
851
+ "environments": [
852
+ "downtown_west_sunset"
853
+ ]
854
+ },
855
+ {
856
+ "file": "syn/train/data-00075-of-00451.parquet",
857
+ "rows": 1000,
858
+ "bytes": 2404777318,
859
+ "classes": {
860
+ "bus_stop": 1000
861
+ },
862
+ "environments": [
863
+ "downtown_west_sunset"
864
+ ]
865
+ },
866
+ {
867
+ "file": "syn/train/data-00076-of-00451.parquet",
868
+ "rows": 1000,
869
+ "bytes": 2343669961,
870
+ "classes": {
871
+ "bus_stop": 1000
872
+ },
873
+ "environments": [
874
+ "downtown_west_sunset",
875
+ "miami_beach"
876
+ ]
877
+ },
878
+ {
879
+ "file": "syn/train/data-00077-of-00451.parquet",
880
+ "rows": 1000,
881
+ "bytes": 2776264449,
882
+ "classes": {
883
+ "bus_stop": 1000
884
+ },
885
+ "environments": [
886
+ "miami_beach"
887
+ ]
888
+ },
889
+ {
890
+ "file": "syn/train/data-00078-of-00451.parquet",
891
+ "rows": 1000,
892
+ "bytes": 2722277655,
893
+ "classes": {
894
+ "bus_stop": 1000
895
+ },
896
+ "environments": [
897
+ "miami_beach"
898
+ ]
899
+ },
900
+ {
901
+ "file": "syn/train/data-00079-of-00451.parquet",
902
+ "rows": 1000,
903
+ "bytes": 2767284671,
904
+ "classes": {
905
+ "bus_stop": 1000
906
+ },
907
+ "environments": [
908
+ "miami_beach"
909
+ ]
910
+ },
911
+ {
912
+ "file": "syn/train/data-00080-of-00451.parquet",
913
+ "rows": 1000,
914
+ "bytes": 2705864908,
915
+ "classes": {
916
+ "bus_stop": 1000
917
+ },
918
+ "environments": [
919
+ "miami_beach",
920
+ "miami_beach_night"
921
+ ]
922
+ },
923
+ {
924
+ "file": "syn/train/data-00081-of-00451.parquet",
925
+ "rows": 1000,
926
+ "bytes": 2720544862,
927
+ "classes": {
928
+ "bus_stop": 1000
929
+ },
930
+ "environments": [
931
+ "miami_beach_night"
932
+ ]
933
+ },
934
+ {
935
+ "file": "syn/train/data-00082-of-00451.parquet",
936
+ "rows": 1000,
937
+ "bytes": 2685557217,
938
+ "classes": {
939
+ "bus_stop": 1000
940
+ },
941
+ "environments": [
942
+ "miami_beach_night"
943
+ ]
944
+ },
945
+ {
946
+ "file": "syn/train/data-00083-of-00451.parquet",
947
+ "rows": 1000,
948
+ "bytes": 2489257708,
949
+ "classes": {
950
+ "bus_stop": 1000
951
+ },
952
+ "environments": [
953
+ "miami_beach_night",
954
+ "neighborhood"
955
+ ]
956
+ },
957
+ {
958
+ "file": "syn/train/data-00084-of-00451.parquet",
959
+ "rows": 1000,
960
+ "bytes": 2073697881,
961
+ "classes": {
962
+ "bus_stop": 1000
963
+ },
964
+ "environments": [
965
+ "neighborhood"
966
+ ]
967
+ },
968
+ {
969
+ "file": "syn/train/data-00085-of-00451.parquet",
970
+ "rows": 1000,
971
+ "bytes": 2130798617,
972
+ "classes": {
973
+ "bus_stop": 1000
974
+ },
975
+ "environments": [
976
+ "neighborhood",
977
+ "neighborhood_sunset"
978
+ ]
979
+ },
980
+ {
981
+ "file": "syn/train/data-00086-of-00451.parquet",
982
+ "rows": 1000,
983
+ "bytes": 2140415174,
984
+ "classes": {
985
+ "bus_stop": 1000
986
+ },
987
+ "environments": [
988
+ "neighborhood_sunset"
989
+ ]
990
+ },
991
+ {
992
+ "file": "syn/train/data-00087-of-00451.parquet",
993
+ "rows": 1000,
994
+ "bytes": 1703250898,
995
+ "classes": {
996
+ "bus_stop": 1000
997
+ },
998
+ "environments": [
999
+ "neighborhood_sunset",
1000
+ "night_city"
1001
+ ]
1002
+ },
1003
+ {
1004
+ "file": "syn/train/data-00088-of-00451.parquet",
1005
+ "rows": 1000,
1006
+ "bytes": 1691538030,
1007
+ "classes": {
1008
+ "bus_stop": 1000
1009
+ },
1010
+ "environments": [
1011
+ "night_city",
1012
+ "small_city"
1013
+ ]
1014
+ },
1015
+ {
1016
+ "file": "syn/train/data-00089-of-00451.parquet",
1017
+ "rows": 1000,
1018
+ "bytes": 1896797341,
1019
+ "classes": {
1020
+ "bus_stop": 1000
1021
+ },
1022
+ "environments": [
1023
+ "small_city"
1024
+ ]
1025
+ },
1026
+ {
1027
+ "file": "syn/train/data-00090-of-00451.parquet",
1028
+ "rows": 1000,
1029
+ "bytes": 1846357990,
1030
+ "classes": {
1031
+ "bus_stop": 1000
1032
+ },
1033
+ "environments": [
1034
+ "small_city"
1035
+ ]
1036
+ },
1037
+ {
1038
+ "file": "syn/train/data-00091-of-00451.parquet",
1039
+ "rows": 1000,
1040
+ "bytes": 1879468054,
1041
+ "classes": {
1042
+ "bus_stop": 1000
1043
+ },
1044
+ "environments": [
1045
+ "small_city",
1046
+ "small_city_sunset"
1047
+ ]
1048
+ },
1049
+ {
1050
+ "file": "syn/train/data-00092-of-00451.parquet",
1051
+ "rows": 1000,
1052
+ "bytes": 1900767052,
1053
+ "classes": {
1054
+ "bus_stop": 1000
1055
+ },
1056
+ "environments": [
1057
+ "small_city_sunset"
1058
+ ]
1059
+ },
1060
+ {
1061
+ "file": "syn/train/data-00093-of-00451.parquet",
1062
+ "rows": 1000,
1063
+ "bytes": 1840477332,
1064
+ "classes": {
1065
+ "bus_stop": 1000
1066
+ },
1067
+ "environments": [
1068
+ "small_city_sunset"
1069
+ ]
1070
+ },
1071
+ {
1072
+ "file": "syn/train/data-00094-of-00451.parquet",
1073
+ "rows": 1000,
1074
+ "bytes": 1867507087,
1075
+ "classes": {
1076
+ "bus_stop": 1000
1077
+ },
1078
+ "environments": [
1079
+ "small_city_sunset",
1080
+ "stylized_city"
1081
+ ]
1082
+ },
1083
+ {
1084
+ "file": "syn/train/data-00095-of-00451.parquet",
1085
+ "rows": 1000,
1086
+ "bytes": 1887800186,
1087
+ "classes": {
1088
+ "bus_stop": 1000
1089
+ },
1090
+ "environments": [
1091
+ "stylized_city"
1092
+ ]
1093
+ },
1094
+ {
1095
+ "file": "syn/train/data-00096-of-00451.parquet",
1096
+ "rows": 1000,
1097
+ "bytes": 1839365485,
1098
+ "classes": {
1099
+ "bus_stop": 1000
1100
+ },
1101
+ "environments": [
1102
+ "stylized_city"
1103
+ ]
1104
+ },
1105
+ {
1106
+ "file": "syn/train/data-00097-of-00451.parquet",
1107
+ "rows": 1000,
1108
+ "bytes": 1767313599,
1109
+ "classes": {
1110
+ "bus_stop": 1000
1111
+ },
1112
+ "environments": [
1113
+ "stylized_city"
1114
+ ]
1115
+ },
1116
+ {
1117
+ "file": "syn/train/data-00098-of-00451.parquet",
1118
+ "rows": 1000,
1119
+ "bytes": 1693847744,
1120
+ "classes": {
1121
+ "bus_stop": 1000
1122
+ },
1123
+ "environments": [
1124
+ "stylized_city",
1125
+ "stylized_city_night"
1126
+ ]
1127
+ },
1128
+ {
1129
+ "file": "syn/train/data-00099-of-00451.parquet",
1130
+ "rows": 1000,
1131
+ "bytes": 1677887907,
1132
+ "classes": {
1133
+ "bus_stop": 1000
1134
+ },
1135
+ "environments": [
1136
+ "stylized_city_night"
1137
+ ]
1138
+ },
1139
+ {
1140
+ "file": "syn/train/data-00100-of-00451.parquet",
1141
+ "rows": 1000,
1142
+ "bytes": 1607408005,
1143
+ "classes": {
1144
+ "bus_stop": 1000
1145
+ },
1146
+ "environments": [
1147
+ "stylized_city_night"
1148
+ ]
1149
+ },
1150
+ {
1151
+ "file": "syn/train/data-00101-of-00451.parquet",
1152
+ "rows": 1000,
1153
+ "bytes": 1489481763,
1154
+ "classes": {
1155
+ "bus_stop": 1000
1156
+ },
1157
+ "environments": [
1158
+ "stylized_city_night"
1159
+ ]
1160
+ },
1161
+ {
1162
+ "file": "syn/train/data-00102-of-00451.parquet",
1163
+ "rows": 1000,
1164
+ "bytes": 1656070705,
1165
+ "classes": {
1166
+ "bus_stop": 1000
1167
+ },
1168
+ "environments": [
1169
+ "stylized_city_night",
1170
+ "stylized_city_sunset"
1171
+ ]
1172
+ },
1173
+ {
1174
+ "file": "syn/train/data-00103-of-00451.parquet",
1175
+ "rows": 1000,
1176
+ "bytes": 1857068427,
1177
+ "classes": {
1178
+ "bus_stop": 1000
1179
+ },
1180
+ "environments": [
1181
+ "stylized_city_sunset"
1182
+ ]
1183
+ },
1184
+ {
1185
+ "file": "syn/train/data-00104-of-00451.parquet",
1186
+ "rows": 1000,
1187
+ "bytes": 1794825892,
1188
+ "classes": {
1189
+ "bus_stop": 1000
1190
+ },
1191
+ "environments": [
1192
+ "stylized_city_sunset"
1193
+ ]
1194
+ },
1195
+ {
1196
+ "file": "syn/train/data-00105-of-00451.parquet",
1197
+ "rows": 1000,
1198
+ "bytes": 1688617156,
1199
+ "classes": {
1200
+ "bus_stop": 1000
1201
+ },
1202
+ "environments": [
1203
+ "stylized_city_sunset"
1204
+ ]
1205
+ },
1206
+ {
1207
+ "file": "syn/train/data-00106-of-00451.parquet",
1208
+ "rows": 1000,
1209
+ "bytes": 1786924039,
1210
+ "classes": {
1211
+ "bus_stop": 1000
1212
+ },
1213
+ "environments": [
1214
+ "stylized_city_sunset",
1215
+ "suburbs"
1216
+ ]
1217
+ },
1218
+ {
1219
+ "file": "syn/train/data-00107-of-00451.parquet",
1220
+ "rows": 1000,
1221
+ "bytes": 1929942675,
1222
+ "classes": {
1223
+ "bus_stop": 1000
1224
+ },
1225
+ "environments": [
1226
+ "suburbs"
1227
+ ]
1228
+ },
1229
+ {
1230
+ "file": "syn/train/data-00108-of-00451.parquet",
1231
+ "rows": 1000,
1232
+ "bytes": 1940017228,
1233
+ "classes": {
1234
+ "bus_stop": 1000
1235
+ },
1236
+ "environments": [
1237
+ "suburbs"
1238
+ ]
1239
+ },
1240
+ {
1241
+ "file": "syn/train/data-00109-of-00451.parquet",
1242
+ "rows": 1000,
1243
+ "bytes": 1759126456,
1244
+ "classes": {
1245
+ "bus_stop": 1000
1246
+ },
1247
+ "environments": [
1248
+ "suburbs"
1249
+ ]
1250
+ },
1251
+ {
1252
+ "file": "syn/train/data-00110-of-00451.parquet",
1253
+ "rows": 1000,
1254
+ "bytes": 1636136112,
1255
+ "classes": {
1256
+ "bus_stop": 1000
1257
+ },
1258
+ "environments": [
1259
+ "suburbs",
1260
+ "suburbs_night"
1261
+ ]
1262
+ },
1263
+ {
1264
+ "file": "syn/train/data-00111-of-00451.parquet",
1265
+ "rows": 1000,
1266
+ "bytes": 1711871341,
1267
+ "classes": {
1268
+ "bus_stop": 1000
1269
+ },
1270
+ "environments": [
1271
+ "suburbs_night"
1272
+ ]
1273
+ },
1274
+ {
1275
+ "file": "syn/train/data-00112-of-00451.parquet",
1276
+ "rows": 1000,
1277
+ "bytes": 1647352922,
1278
+ "classes": {
1279
+ "bus_stop": 1000
1280
+ },
1281
+ "environments": [
1282
+ "suburbs_night"
1283
+ ]
1284
+ },
1285
+ {
1286
+ "file": "syn/train/data-00113-of-00451.parquet",
1287
+ "rows": 1000,
1288
+ "bytes": 1522502121,
1289
+ "classes": {
1290
+ "bus_stop": 1000
1291
+ },
1292
+ "environments": [
1293
+ "suburbs_night"
1294
+ ]
1295
+ },
1296
+ {
1297
+ "file": "syn/train/data-00114-of-00451.parquet",
1298
+ "rows": 1000,
1299
+ "bytes": 1681604303,
1300
+ "classes": {
1301
+ "bus_stop": 1000
1302
+ },
1303
+ "environments": [
1304
+ "suburbs_night",
1305
+ "suburbs_sunset"
1306
+ ]
1307
+ },
1308
+ {
1309
+ "file": "syn/train/data-00115-of-00451.parquet",
1310
+ "rows": 1000,
1311
+ "bytes": 1836195364,
1312
+ "classes": {
1313
+ "bus_stop": 1000
1314
+ },
1315
+ "environments": [
1316
+ "suburbs_sunset"
1317
+ ]
1318
+ },
1319
+ {
1320
+ "file": "syn/train/data-00116-of-00451.parquet",
1321
+ "rows": 1000,
1322
+ "bytes": 1878040895,
1323
+ "classes": {
1324
+ "bus_stop": 1000
1325
+ },
1326
+ "environments": [
1327
+ "suburbs_sunset"
1328
+ ]
1329
+ },
1330
+ {
1331
+ "file": "syn/train/data-00117-of-00451.parquet",
1332
+ "rows": 1000,
1333
+ "bytes": 1734942751,
1334
+ "classes": {
1335
+ "bus_stop": 1000
1336
+ },
1337
+ "environments": [
1338
+ "suburbs_sunset"
1339
+ ]
1340
+ },
1341
+ {
1342
+ "file": "syn/train/data-00118-of-00451.parquet",
1343
+ "rows": 1000,
1344
+ "bytes": 1800093749,
1345
+ "classes": {
1346
+ "bus_stop": 1000
1347
+ },
1348
+ "environments": [
1349
+ "suburbs_sunset",
1350
+ "village"
1351
+ ]
1352
+ },
1353
+ {
1354
+ "file": "syn/train/data-00119-of-00451.parquet",
1355
+ "rows": 1000,
1356
+ "bytes": 1819300914,
1357
+ "classes": {
1358
+ "bus_stop": 1000
1359
+ },
1360
+ "environments": [
1361
+ "village"
1362
+ ]
1363
+ },
1364
+ {
1365
+ "file": "syn/train/data-00120-of-00451.parquet",
1366
+ "rows": 1000,
1367
+ "bytes": 1835465314,
1368
+ "classes": {
1369
+ "bus_stop": 1000
1370
+ },
1371
+ "environments": [
1372
+ "village",
1373
+ "village_sunset"
1374
+ ]
1375
+ },
1376
+ {
1377
+ "file": "syn/train/data-00121-of-00451.parquet",
1378
+ "rows": 1000,
1379
+ "bytes": 1868543032,
1380
+ "classes": {
1381
+ "bus_stop": 1000
1382
+ },
1383
+ "environments": [
1384
+ "village_sunset"
1385
+ ]
1386
+ },
1387
+ {
1388
+ "file": "syn/train/data-00122-of-00451.parquet",
1389
+ "rows": 1000,
1390
+ "bytes": 1854386606,
1391
+ "classes": {
1392
+ "bus_stop": 1000
1393
+ },
1394
+ "environments": [
1395
+ "village_sunset"
1396
+ ]
1397
+ },
1398
+ {
1399
+ "file": "syn/train/data-00123-of-00451.parquet",
1400
+ "rows": 1000,
1401
+ "bytes": 1842625563,
1402
+ "classes": {
1403
+ "bus_stop": 460,
1404
+ "bus_stop_sign": 540
1405
+ },
1406
+ "environments": [
1407
+ "downtown_west",
1408
+ "village_sunset"
1409
+ ]
1410
+ },
1411
+ {
1412
+ "file": "syn/train/data-00124-of-00451.parquet",
1413
+ "rows": 1000,
1414
+ "bytes": 1884448865,
1415
+ "classes": {
1416
+ "bus_stop_sign": 1000
1417
+ },
1418
+ "environments": [
1419
+ "downtown_west"
1420
+ ]
1421
+ },
1422
+ {
1423
+ "file": "syn/train/data-00125-of-00451.parquet",
1424
+ "rows": 1000,
1425
+ "bytes": 1915694207,
1426
+ "classes": {
1427
+ "bus_stop_sign": 1000
1428
+ },
1429
+ "environments": [
1430
+ "downtown_west"
1431
+ ]
1432
+ },
1433
+ {
1434
+ "file": "syn/train/data-00126-of-00451.parquet",
1435
+ "rows": 1000,
1436
+ "bytes": 2087578048,
1437
+ "classes": {
1438
+ "bus_stop_sign": 1000
1439
+ },
1440
+ "environments": [
1441
+ "downtown_west"
1442
+ ]
1443
+ },
1444
+ {
1445
+ "file": "syn/train/data-00127-of-00451.parquet",
1446
+ "rows": 1000,
1447
+ "bytes": 2374932502,
1448
+ "classes": {
1449
+ "bus_stop_sign": 1000
1450
+ },
1451
+ "environments": [
1452
+ "downtown_west"
1453
+ ]
1454
+ },
1455
+ {
1456
+ "file": "syn/train/data-00128-of-00451.parquet",
1457
+ "rows": 1000,
1458
+ "bytes": 1610072197,
1459
+ "classes": {
1460
+ "bus_stop_sign": 1000
1461
+ },
1462
+ "environments": [
1463
+ "downtown_west",
1464
+ "downtown_west_night"
1465
+ ]
1466
+ },
1467
+ {
1468
+ "file": "syn/train/data-00129-of-00451.parquet",
1469
+ "rows": 1000,
1470
+ "bytes": 1581628956,
1471
+ "classes": {
1472
+ "bus_stop_sign": 1000
1473
+ },
1474
+ "environments": [
1475
+ "downtown_west_night"
1476
+ ]
1477
+ },
1478
+ {
1479
+ "file": "syn/train/data-00130-of-00451.parquet",
1480
+ "rows": 1000,
1481
+ "bytes": 1446574909,
1482
+ "classes": {
1483
+ "bus_stop_sign": 1000
1484
+ },
1485
+ "environments": [
1486
+ "downtown_west_night"
1487
+ ]
1488
+ },
1489
+ {
1490
+ "file": "syn/train/data-00131-of-00451.parquet",
1491
+ "rows": 1000,
1492
+ "bytes": 1911466323,
1493
+ "classes": {
1494
+ "bus_stop_sign": 1000
1495
+ },
1496
+ "environments": [
1497
+ "downtown_west_night"
1498
+ ]
1499
+ },
1500
+ {
1501
+ "file": "syn/train/data-00132-of-00451.parquet",
1502
+ "rows": 1000,
1503
+ "bytes": 1859369714,
1504
+ "classes": {
1505
+ "bus_stop_sign": 1000
1506
+ },
1507
+ "environments": [
1508
+ "downtown_west_night",
1509
+ "downtown_west_sunset"
1510
+ ]
1511
+ },
1512
+ {
1513
+ "file": "syn/train/data-00133-of-00451.parquet",
1514
+ "rows": 1000,
1515
+ "bytes": 1793983267,
1516
+ "classes": {
1517
+ "bus_stop_sign": 1000
1518
+ },
1519
+ "environments": [
1520
+ "downtown_west_sunset"
1521
+ ]
1522
+ },
1523
+ {
1524
+ "file": "syn/train/data-00134-of-00451.parquet",
1525
+ "rows": 1000,
1526
+ "bytes": 1910150429,
1527
+ "classes": {
1528
+ "bus_stop_sign": 1000
1529
+ },
1530
+ "environments": [
1531
+ "downtown_west_sunset"
1532
+ ]
1533
+ },
1534
+ {
1535
+ "file": "syn/train/data-00135-of-00451.parquet",
1536
+ "rows": 1000,
1537
+ "bytes": 1612356316,
1538
+ "classes": {
1539
+ "bus_stop_sign": 1000
1540
+ },
1541
+ "environments": [
1542
+ "downtown_west_sunset"
1543
+ ]
1544
+ },
1545
+ {
1546
+ "file": "syn/train/data-00136-of-00451.parquet",
1547
+ "rows": 1000,
1548
+ "bytes": 2519103508,
1549
+ "classes": {
1550
+ "bus_stop_sign": 1000
1551
+ },
1552
+ "environments": [
1553
+ "downtown_west_sunset"
1554
+ ]
1555
+ },
1556
+ {
1557
+ "file": "syn/train/data-00137-of-00451.parquet",
1558
+ "rows": 1000,
1559
+ "bytes": 2411967422,
1560
+ "classes": {
1561
+ "bus_stop_sign": 1000
1562
+ },
1563
+ "environments": [
1564
+ "downtown_west_sunset",
1565
+ "miami_beach"
1566
+ ]
1567
+ },
1568
+ {
1569
+ "file": "syn/train/data-00138-of-00451.parquet",
1570
+ "rows": 1000,
1571
+ "bytes": 2904141167,
1572
+ "classes": {
1573
+ "bus_stop_sign": 1000
1574
+ },
1575
+ "environments": [
1576
+ "miami_beach"
1577
+ ]
1578
+ },
1579
+ {
1580
+ "file": "syn/train/data-00139-of-00451.parquet",
1581
+ "rows": 1000,
1582
+ "bytes": 2754233501,
1583
+ "classes": {
1584
+ "bus_stop_sign": 1000
1585
+ },
1586
+ "environments": [
1587
+ "miami_beach"
1588
+ ]
1589
+ },
1590
+ {
1591
+ "file": "syn/train/data-00140-of-00451.parquet",
1592
+ "rows": 1000,
1593
+ "bytes": 2777072211,
1594
+ "classes": {
1595
+ "bus_stop_sign": 1000
1596
+ },
1597
+ "environments": [
1598
+ "miami_beach"
1599
+ ]
1600
+ },
1601
+ {
1602
+ "file": "syn/train/data-00141-of-00451.parquet",
1603
+ "rows": 1000,
1604
+ "bytes": 2874063552,
1605
+ "classes": {
1606
+ "bus_stop_sign": 1000
1607
+ },
1608
+ "environments": [
1609
+ "miami_beach",
1610
+ "miami_beach_night"
1611
+ ]
1612
+ },
1613
+ {
1614
+ "file": "syn/train/data-00142-of-00451.parquet",
1615
+ "rows": 1000,
1616
+ "bytes": 2828844628,
1617
+ "classes": {
1618
+ "bus_stop_sign": 1000
1619
+ },
1620
+ "environments": [
1621
+ "miami_beach_night"
1622
+ ]
1623
+ },
1624
+ {
1625
+ "file": "syn/train/data-00143-of-00451.parquet",
1626
+ "rows": 1000,
1627
+ "bytes": 2779987490,
1628
+ "classes": {
1629
+ "bus_stop_sign": 1000
1630
+ },
1631
+ "environments": [
1632
+ "miami_beach_night"
1633
+ ]
1634
+ },
1635
+ {
1636
+ "file": "syn/train/data-00144-of-00451.parquet",
1637
+ "rows": 1000,
1638
+ "bytes": 2361856536,
1639
+ "classes": {
1640
+ "bus_stop_sign": 1000
1641
+ },
1642
+ "environments": [
1643
+ "miami_beach_night",
1644
+ "neighborhood"
1645
+ ]
1646
+ },
1647
+ {
1648
+ "file": "syn/train/data-00145-of-00451.parquet",
1649
+ "rows": 1000,
1650
+ "bytes": 2085875725,
1651
+ "classes": {
1652
+ "bus_stop_sign": 1000
1653
+ },
1654
+ "environments": [
1655
+ "neighborhood"
1656
+ ]
1657
+ },
1658
+ {
1659
+ "file": "syn/train/data-00146-of-00451.parquet",
1660
+ "rows": 1000,
1661
+ "bytes": 2119021365,
1662
+ "classes": {
1663
+ "bus_stop_sign": 1000
1664
+ },
1665
+ "environments": [
1666
+ "neighborhood",
1667
+ "neighborhood_sunset"
1668
+ ]
1669
+ },
1670
+ {
1671
+ "file": "syn/train/data-00147-of-00451.parquet",
1672
+ "rows": 1000,
1673
+ "bytes": 2110176473,
1674
+ "classes": {
1675
+ "bus_stop_sign": 1000
1676
+ },
1677
+ "environments": [
1678
+ "neighborhood_sunset",
1679
+ "night_city"
1680
+ ]
1681
+ },
1682
+ {
1683
+ "file": "syn/train/data-00148-of-00451.parquet",
1684
+ "rows": 1000,
1685
+ "bytes": 1552615688,
1686
+ "classes": {
1687
+ "bus_stop_sign": 1000
1688
+ },
1689
+ "environments": [
1690
+ "night_city"
1691
+ ]
1692
+ },
1693
+ {
1694
+ "file": "syn/train/data-00149-of-00451.parquet",
1695
+ "rows": 1000,
1696
+ "bytes": 1732493108,
1697
+ "classes": {
1698
+ "bus_stop_sign": 1000
1699
+ },
1700
+ "environments": [
1701
+ "night_city",
1702
+ "small_city"
1703
+ ]
1704
+ },
1705
+ {
1706
+ "file": "syn/train/data-00150-of-00451.parquet",
1707
+ "rows": 1000,
1708
+ "bytes": 1897664693,
1709
+ "classes": {
1710
+ "bus_stop_sign": 1000
1711
+ },
1712
+ "environments": [
1713
+ "small_city"
1714
+ ]
1715
+ },
1716
+ {
1717
+ "file": "syn/train/data-00151-of-00451.parquet",
1718
+ "rows": 1000,
1719
+ "bytes": 1856792038,
1720
+ "classes": {
1721
+ "bus_stop_sign": 1000
1722
+ },
1723
+ "environments": [
1724
+ "small_city"
1725
+ ]
1726
+ },
1727
+ {
1728
+ "file": "syn/train/data-00152-of-00451.parquet",
1729
+ "rows": 1000,
1730
+ "bytes": 1897001062,
1731
+ "classes": {
1732
+ "bus_stop_sign": 1000
1733
+ },
1734
+ "environments": [
1735
+ "small_city",
1736
+ "small_city_sunset"
1737
+ ]
1738
+ },
1739
+ {
1740
+ "file": "syn/train/data-00153-of-00451.parquet",
1741
+ "rows": 1000,
1742
+ "bytes": 1887150139,
1743
+ "classes": {
1744
+ "bus_stop_sign": 1000
1745
+ },
1746
+ "environments": [
1747
+ "small_city_sunset"
1748
+ ]
1749
+ },
1750
+ {
1751
+ "file": "syn/train/data-00154-of-00451.parquet",
1752
+ "rows": 1000,
1753
+ "bytes": 1852558634,
1754
+ "classes": {
1755
+ "bus_stop_sign": 1000
1756
+ },
1757
+ "environments": [
1758
+ "small_city_sunset"
1759
+ ]
1760
+ },
1761
+ {
1762
+ "file": "syn/train/data-00155-of-00451.parquet",
1763
+ "rows": 1000,
1764
+ "bytes": 1898399830,
1765
+ "classes": {
1766
+ "bus_stop_sign": 1000
1767
+ },
1768
+ "environments": [
1769
+ "small_city_sunset",
1770
+ "stylized_city"
1771
+ ]
1772
+ },
1773
+ {
1774
+ "file": "syn/train/data-00156-of-00451.parquet",
1775
+ "rows": 1000,
1776
+ "bytes": 1920708417,
1777
+ "classes": {
1778
+ "bus_stop_sign": 1000
1779
+ },
1780
+ "environments": [
1781
+ "stylized_city"
1782
+ ]
1783
+ },
1784
+ {
1785
+ "file": "syn/train/data-00157-of-00451.parquet",
1786
+ "rows": 1000,
1787
+ "bytes": 1777128893,
1788
+ "classes": {
1789
+ "bus_stop_sign": 1000
1790
+ },
1791
+ "environments": [
1792
+ "stylized_city"
1793
+ ]
1794
+ },
1795
+ {
1796
+ "file": "syn/train/data-00158-of-00451.parquet",
1797
+ "rows": 1000,
1798
+ "bytes": 1734373288,
1799
+ "classes": {
1800
+ "bus_stop_sign": 1000
1801
+ },
1802
+ "environments": [
1803
+ "stylized_city"
1804
+ ]
1805
+ },
1806
+ {
1807
+ "file": "syn/train/data-00159-of-00451.parquet",
1808
+ "rows": 1000,
1809
+ "bytes": 1720188205,
1810
+ "classes": {
1811
+ "bus_stop_sign": 1000
1812
+ },
1813
+ "environments": [
1814
+ "stylized_city",
1815
+ "stylized_city_night"
1816
+ ]
1817
+ },
1818
+ {
1819
+ "file": "syn/train/data-00160-of-00451.parquet",
1820
+ "rows": 1000,
1821
+ "bytes": 1685565492,
1822
+ "classes": {
1823
+ "bus_stop_sign": 1000
1824
+ },
1825
+ "environments": [
1826
+ "stylized_city_night"
1827
+ ]
1828
+ },
1829
+ {
1830
+ "file": "syn/train/data-00161-of-00451.parquet",
1831
+ "rows": 1000,
1832
+ "bytes": 1513839400,
1833
+ "classes": {
1834
+ "bus_stop_sign": 1000
1835
+ },
1836
+ "environments": [
1837
+ "stylized_city_night"
1838
+ ]
1839
+ },
1840
+ {
1841
+ "file": "syn/train/data-00162-of-00451.parquet",
1842
+ "rows": 1000,
1843
+ "bytes": 1498657910,
1844
+ "classes": {
1845
+ "bus_stop_sign": 1000
1846
+ },
1847
+ "environments": [
1848
+ "stylized_city_night",
1849
+ "stylized_city_sunset"
1850
+ ]
1851
+ },
1852
+ {
1853
+ "file": "syn/train/data-00163-of-00451.parquet",
1854
+ "rows": 1000,
1855
+ "bytes": 1895061013,
1856
+ "classes": {
1857
+ "bus_stop_sign": 1000
1858
+ },
1859
+ "environments": [
1860
+ "stylized_city_sunset"
1861
+ ]
1862
+ },
1863
+ {
1864
+ "file": "syn/train/data-00164-of-00451.parquet",
1865
+ "rows": 1000,
1866
+ "bytes": 1837532419,
1867
+ "classes": {
1868
+ "bus_stop_sign": 1000
1869
+ },
1870
+ "environments": [
1871
+ "stylized_city_sunset"
1872
+ ]
1873
+ },
1874
+ {
1875
+ "file": "syn/train/data-00165-of-00451.parquet",
1876
+ "rows": 1000,
1877
+ "bytes": 1729938002,
1878
+ "classes": {
1879
+ "bus_stop_sign": 1000
1880
+ },
1881
+ "environments": [
1882
+ "stylized_city_sunset"
1883
+ ]
1884
+ },
1885
+ {
1886
+ "file": "syn/train/data-00166-of-00451.parquet",
1887
+ "rows": 1000,
1888
+ "bytes": 1752484628,
1889
+ "classes": {
1890
+ "bus_stop_sign": 1000
1891
+ },
1892
+ "environments": [
1893
+ "stylized_city_sunset",
1894
+ "suburbs"
1895
+ ]
1896
+ },
1897
+ {
1898
+ "file": "syn/train/data-00167-of-00451.parquet",
1899
+ "rows": 1000,
1900
+ "bytes": 1848652705,
1901
+ "classes": {
1902
+ "bus_stop_sign": 1000
1903
+ },
1904
+ "environments": [
1905
+ "suburbs"
1906
+ ]
1907
+ },
1908
+ {
1909
+ "file": "syn/train/data-00168-of-00451.parquet",
1910
+ "rows": 1000,
1911
+ "bytes": 2029089334,
1912
+ "classes": {
1913
+ "bus_stop_sign": 1000
1914
+ },
1915
+ "environments": [
1916
+ "suburbs"
1917
+ ]
1918
+ },
1919
+ {
1920
+ "file": "syn/train/data-00169-of-00451.parquet",
1921
+ "rows": 1000,
1922
+ "bytes": 1860010421,
1923
+ "classes": {
1924
+ "bus_stop_sign": 1000
1925
+ },
1926
+ "environments": [
1927
+ "suburbs"
1928
+ ]
1929
+ },
1930
+ {
1931
+ "file": "syn/train/data-00170-of-00451.parquet",
1932
+ "rows": 1000,
1933
+ "bytes": 1701280337,
1934
+ "classes": {
1935
+ "bus_stop_sign": 1000
1936
+ },
1937
+ "environments": [
1938
+ "suburbs",
1939
+ "suburbs_night"
1940
+ ]
1941
+ },
1942
+ {
1943
+ "file": "syn/train/data-00171-of-00451.parquet",
1944
+ "rows": 1000,
1945
+ "bytes": 1669609087,
1946
+ "classes": {
1947
+ "bus_stop_sign": 1000
1948
+ },
1949
+ "environments": [
1950
+ "suburbs_night"
1951
+ ]
1952
+ },
1953
+ {
1954
+ "file": "syn/train/data-00172-of-00451.parquet",
1955
+ "rows": 1000,
1956
+ "bytes": 1788286246,
1957
+ "classes": {
1958
+ "bus_stop_sign": 1000
1959
+ },
1960
+ "environments": [
1961
+ "suburbs_night"
1962
+ ]
1963
+ },
1964
+ {
1965
+ "file": "syn/train/data-00173-of-00451.parquet",
1966
+ "rows": 1000,
1967
+ "bytes": 1567446992,
1968
+ "classes": {
1969
+ "bus_stop_sign": 1000
1970
+ },
1971
+ "environments": [
1972
+ "suburbs_night"
1973
+ ]
1974
+ },
1975
+ {
1976
+ "file": "syn/train/data-00174-of-00451.parquet",
1977
+ "rows": 1000,
1978
+ "bytes": 1578511789,
1979
+ "classes": {
1980
+ "bus_stop_sign": 1000
1981
+ },
1982
+ "environments": [
1983
+ "suburbs_night",
1984
+ "suburbs_sunset"
1985
+ ]
1986
+ },
1987
+ {
1988
+ "file": "syn/train/data-00175-of-00451.parquet",
1989
+ "rows": 1000,
1990
+ "bytes": 1758624672,
1991
+ "classes": {
1992
+ "bus_stop_sign": 1000
1993
+ },
1994
+ "environments": [
1995
+ "suburbs_sunset"
1996
+ ]
1997
+ },
1998
+ {
1999
+ "file": "syn/train/data-00176-of-00451.parquet",
2000
+ "rows": 1000,
2001
+ "bytes": 1968566395,
2002
+ "classes": {
2003
+ "bus_stop_sign": 1000
2004
+ },
2005
+ "environments": [
2006
+ "suburbs_sunset"
2007
+ ]
2008
+ },
2009
+ {
2010
+ "file": "syn/train/data-00177-of-00451.parquet",
2011
+ "rows": 1000,
2012
+ "bytes": 1810069332,
2013
+ "classes": {
2014
+ "bus_stop_sign": 1000
2015
+ },
2016
+ "environments": [
2017
+ "suburbs_sunset"
2018
+ ]
2019
+ },
2020
+ {
2021
+ "file": "syn/train/data-00178-of-00451.parquet",
2022
+ "rows": 1000,
2023
+ "bytes": 1709346849,
2024
+ "classes": {
2025
+ "bus_stop_sign": 1000
2026
+ },
2027
+ "environments": [
2028
+ "suburbs_sunset",
2029
+ "village"
2030
+ ]
2031
+ },
2032
+ {
2033
+ "file": "syn/train/data-00179-of-00451.parquet",
2034
+ "rows": 1000,
2035
+ "bytes": 1763653947,
2036
+ "classes": {
2037
+ "bus_stop_sign": 1000
2038
+ },
2039
+ "environments": [
2040
+ "village"
2041
+ ]
2042
+ },
2043
+ {
2044
+ "file": "syn/train/data-00180-of-00451.parquet",
2045
+ "rows": 1000,
2046
+ "bytes": 1755627952,
2047
+ "classes": {
2048
+ "bus_stop_sign": 1000
2049
+ },
2050
+ "environments": [
2051
+ "village"
2052
+ ]
2053
+ },
2054
+ {
2055
+ "file": "syn/train/data-00181-of-00451.parquet",
2056
+ "rows": 1000,
2057
+ "bytes": 1786846478,
2058
+ "classes": {
2059
+ "bus_stop_sign": 1000
2060
+ },
2061
+ "environments": [
2062
+ "village",
2063
+ "village_sunset"
2064
+ ]
2065
+ },
2066
+ {
2067
+ "file": "syn/train/data-00182-of-00451.parquet",
2068
+ "rows": 1000,
2069
+ "bytes": 1782146053,
2070
+ "classes": {
2071
+ "bus_stop_sign": 1000
2072
+ },
2073
+ "environments": [
2074
+ "village_sunset"
2075
+ ]
2076
+ },
2077
+ {
2078
+ "file": "syn/train/data-00183-of-00451.parquet",
2079
+ "rows": 1000,
2080
+ "bytes": 1808141418,
2081
+ "classes": {
2082
+ "bus_stop_sign": 910,
2083
+ "crosswalk": 90
2084
+ },
2085
+ "environments": [
2086
+ "downtown_west",
2087
+ "village_sunset"
2088
+ ]
2089
+ },
2090
+ {
2091
+ "file": "syn/train/data-00184-of-00451.parquet",
2092
+ "rows": 1000,
2093
+ "bytes": 1783786274,
2094
+ "classes": {
2095
+ "crosswalk": 1000
2096
+ },
2097
+ "environments": [
2098
+ "downtown_west"
2099
+ ]
2100
+ },
2101
+ {
2102
+ "file": "syn/train/data-00185-of-00451.parquet",
2103
+ "rows": 1000,
2104
+ "bytes": 1650421583,
2105
+ "classes": {
2106
+ "crosswalk": 1000
2107
+ },
2108
+ "environments": [
2109
+ "downtown_west"
2110
+ ]
2111
+ },
2112
+ {
2113
+ "file": "syn/train/data-00186-of-00451.parquet",
2114
+ "rows": 1000,
2115
+ "bytes": 1555322220,
2116
+ "classes": {
2117
+ "crosswalk": 1000
2118
+ },
2119
+ "environments": [
2120
+ "downtown_west",
2121
+ "downtown_west_night"
2122
+ ]
2123
+ },
2124
+ {
2125
+ "file": "syn/train/data-00187-of-00451.parquet",
2126
+ "rows": 1000,
2127
+ "bytes": 1441826684,
2128
+ "classes": {
2129
+ "crosswalk": 1000
2130
+ },
2131
+ "environments": [
2132
+ "downtown_west_night"
2133
+ ]
2134
+ },
2135
+ {
2136
+ "file": "syn/train/data-00188-of-00451.parquet",
2137
+ "rows": 1000,
2138
+ "bytes": 1383684980,
2139
+ "classes": {
2140
+ "crosswalk": 1000
2141
+ },
2142
+ "environments": [
2143
+ "downtown_west_night"
2144
+ ]
2145
+ },
2146
+ {
2147
+ "file": "syn/train/data-00189-of-00451.parquet",
2148
+ "rows": 1000,
2149
+ "bytes": 1709207526,
2150
+ "classes": {
2151
+ "crosswalk": 1000
2152
+ },
2153
+ "environments": [
2154
+ "downtown_west_night",
2155
+ "downtown_west_sunset"
2156
+ ]
2157
+ },
2158
+ {
2159
+ "file": "syn/train/data-00190-of-00451.parquet",
2160
+ "rows": 1000,
2161
+ "bytes": 1594637887,
2162
+ "classes": {
2163
+ "crosswalk": 1000
2164
+ },
2165
+ "environments": [
2166
+ "downtown_west_sunset"
2167
+ ]
2168
+ },
2169
+ {
2170
+ "file": "syn/train/data-00191-of-00451.parquet",
2171
+ "rows": 1000,
2172
+ "bytes": 1664839088,
2173
+ "classes": {
2174
+ "crosswalk": 1000
2175
+ },
2176
+ "environments": [
2177
+ "downtown_west_sunset",
2178
+ "hong_kong_street"
2179
+ ]
2180
+ },
2181
+ {
2182
+ "file": "syn/train/data-00192-of-00451.parquet",
2183
+ "rows": 1000,
2184
+ "bytes": 1812045401,
2185
+ "classes": {
2186
+ "crosswalk": 1000
2187
+ },
2188
+ "environments": [
2189
+ "hong_kong_street",
2190
+ "hong_kong_street_sunset"
2191
+ ]
2192
+ },
2193
+ {
2194
+ "file": "syn/train/data-00193-of-00451.parquet",
2195
+ "rows": 1000,
2196
+ "bytes": 1953594325,
2197
+ "classes": {
2198
+ "crosswalk": 1000
2199
+ },
2200
+ "environments": [
2201
+ "hong_kong_street_sunset",
2202
+ "miami_beach"
2203
+ ]
2204
+ },
2205
+ {
2206
+ "file": "syn/train/data-00194-of-00451.parquet",
2207
+ "rows": 1000,
2208
+ "bytes": 2231432064,
2209
+ "classes": {
2210
+ "crosswalk": 1000
2211
+ },
2212
+ "environments": [
2213
+ "miami_beach"
2214
+ ]
2215
+ },
2216
+ {
2217
+ "file": "syn/train/data-00195-of-00451.parquet",
2218
+ "rows": 1000,
2219
+ "bytes": 2685935203,
2220
+ "classes": {
2221
+ "crosswalk": 1000
2222
+ },
2223
+ "environments": [
2224
+ "miami_beach"
2225
+ ]
2226
+ },
2227
+ {
2228
+ "file": "syn/train/data-00196-of-00451.parquet",
2229
+ "rows": 1000,
2230
+ "bytes": 2800172716,
2231
+ "classes": {
2232
+ "crosswalk": 1000
2233
+ },
2234
+ "environments": [
2235
+ "miami_beach"
2236
+ ]
2237
+ },
2238
+ {
2239
+ "file": "syn/train/data-00197-of-00451.parquet",
2240
+ "rows": 1000,
2241
+ "bytes": 2751509866,
2242
+ "classes": {
2243
+ "crosswalk": 1000
2244
+ },
2245
+ "environments": [
2246
+ "miami_beach",
2247
+ "miami_beach_night"
2248
+ ]
2249
+ },
2250
+ {
2251
+ "file": "syn/train/data-00198-of-00451.parquet",
2252
+ "rows": 1000,
2253
+ "bytes": 2155363337,
2254
+ "classes": {
2255
+ "crosswalk": 1000
2256
+ },
2257
+ "environments": [
2258
+ "miami_beach_night"
2259
+ ]
2260
+ },
2261
+ {
2262
+ "file": "syn/train/data-00199-of-00451.parquet",
2263
+ "rows": 1000,
2264
+ "bytes": 2604169419,
2265
+ "classes": {
2266
+ "crosswalk": 1000
2267
+ },
2268
+ "environments": [
2269
+ "miami_beach_night"
2270
+ ]
2271
+ },
2272
+ {
2273
+ "file": "syn/train/data-00200-of-00451.parquet",
2274
+ "rows": 1000,
2275
+ "bytes": 2752405730,
2276
+ "classes": {
2277
+ "crosswalk": 1000
2278
+ },
2279
+ "environments": [
2280
+ "miami_beach_night"
2281
+ ]
2282
+ },
2283
+ {
2284
+ "file": "syn/train/data-00201-of-00451.parquet",
2285
+ "rows": 1000,
2286
+ "bytes": 2840686083,
2287
+ "classes": {
2288
+ "crosswalk": 1000
2289
+ },
2290
+ "environments": [
2291
+ "miami_beach_night",
2292
+ "night_city"
2293
+ ]
2294
+ },
2295
+ {
2296
+ "file": "syn/train/data-00202-of-00451.parquet",
2297
+ "rows": 1000,
2298
+ "bytes": 1847832412,
2299
+ "classes": {
2300
+ "crosswalk": 1000
2301
+ },
2302
+ "environments": [
2303
+ "night_city"
2304
+ ]
2305
+ },
2306
+ {
2307
+ "file": "syn/train/data-00203-of-00451.parquet",
2308
+ "rows": 1000,
2309
+ "bytes": 1838947576,
2310
+ "classes": {
2311
+ "crosswalk": 1000
2312
+ },
2313
+ "environments": [
2314
+ "night_city"
2315
+ ]
2316
+ },
2317
+ {
2318
+ "file": "syn/train/data-00204-of-00451.parquet",
2319
+ "rows": 1000,
2320
+ "bytes": 1803292679,
2321
+ "classes": {
2322
+ "crosswalk": 1000
2323
+ },
2324
+ "environments": [
2325
+ "night_city",
2326
+ "small_city"
2327
+ ]
2328
+ },
2329
+ {
2330
+ "file": "syn/train/data-00205-of-00451.parquet",
2331
+ "rows": 1000,
2332
+ "bytes": 1816008020,
2333
+ "classes": {
2334
+ "crosswalk": 1000
2335
+ },
2336
+ "environments": [
2337
+ "small_city"
2338
+ ]
2339
+ },
2340
+ {
2341
+ "file": "syn/train/data-00206-of-00451.parquet",
2342
+ "rows": 1000,
2343
+ "bytes": 1783570845,
2344
+ "classes": {
2345
+ "crosswalk": 1000
2346
+ },
2347
+ "environments": [
2348
+ "small_city",
2349
+ "small_city_sunset"
2350
+ ]
2351
+ },
2352
+ {
2353
+ "file": "syn/train/data-00207-of-00451.parquet",
2354
+ "rows": 1000,
2355
+ "bytes": 1866452236,
2356
+ "classes": {
2357
+ "crosswalk": 1000
2358
+ },
2359
+ "environments": [
2360
+ "small_city_sunset"
2361
+ ]
2362
+ },
2363
+ {
2364
+ "file": "syn/train/data-00208-of-00451.parquet",
2365
+ "rows": 1000,
2366
+ "bytes": 1856836510,
2367
+ "classes": {
2368
+ "crosswalk": 1000
2369
+ },
2370
+ "environments": [
2371
+ "small_city_sunset"
2372
+ ]
2373
+ },
2374
+ {
2375
+ "file": "syn/train/data-00209-of-00451.parquet",
2376
+ "rows": 1000,
2377
+ "bytes": 1847962291,
2378
+ "classes": {
2379
+ "crosswalk": 1000
2380
+ },
2381
+ "environments": [
2382
+ "small_city_sunset",
2383
+ "stylized_city"
2384
+ ]
2385
+ },
2386
+ {
2387
+ "file": "syn/train/data-00210-of-00451.parquet",
2388
+ "rows": 1000,
2389
+ "bytes": 1965900842,
2390
+ "classes": {
2391
+ "crosswalk": 1000
2392
+ },
2393
+ "environments": [
2394
+ "stylized_city"
2395
+ ]
2396
+ },
2397
+ {
2398
+ "file": "syn/train/data-00211-of-00451.parquet",
2399
+ "rows": 1000,
2400
+ "bytes": 1748244824,
2401
+ "classes": {
2402
+ "crosswalk": 1000
2403
+ },
2404
+ "environments": [
2405
+ "stylized_city"
2406
+ ]
2407
+ },
2408
+ {
2409
+ "file": "syn/train/data-00212-of-00451.parquet",
2410
+ "rows": 1000,
2411
+ "bytes": 1845121872,
2412
+ "classes": {
2413
+ "crosswalk": 1000
2414
+ },
2415
+ "environments": [
2416
+ "stylized_city"
2417
+ ]
2418
+ },
2419
+ {
2420
+ "file": "syn/train/data-00213-of-00451.parquet",
2421
+ "rows": 1000,
2422
+ "bytes": 1700994107,
2423
+ "classes": {
2424
+ "crosswalk": 1000
2425
+ },
2426
+ "environments": [
2427
+ "stylized_city",
2428
+ "stylized_city_night"
2429
+ ]
2430
+ },
2431
+ {
2432
+ "file": "syn/train/data-00214-of-00451.parquet",
2433
+ "rows": 1000,
2434
+ "bytes": 1696434623,
2435
+ "classes": {
2436
+ "crosswalk": 1000
2437
+ },
2438
+ "environments": [
2439
+ "stylized_city_night"
2440
+ ]
2441
+ },
2442
+ {
2443
+ "file": "syn/train/data-00215-of-00451.parquet",
2444
+ "rows": 1000,
2445
+ "bytes": 1616331612,
2446
+ "classes": {
2447
+ "crosswalk": 1000
2448
+ },
2449
+ "environments": [
2450
+ "stylized_city_night"
2451
+ ]
2452
+ },
2453
+ {
2454
+ "file": "syn/train/data-00216-of-00451.parquet",
2455
+ "rows": 1000,
2456
+ "bytes": 1632741237,
2457
+ "classes": {
2458
+ "crosswalk": 1000
2459
+ },
2460
+ "environments": [
2461
+ "stylized_city_night"
2462
+ ]
2463
+ },
2464
+ {
2465
+ "file": "syn/train/data-00217-of-00451.parquet",
2466
+ "rows": 1000,
2467
+ "bytes": 1760092801,
2468
+ "classes": {
2469
+ "crosswalk": 1000
2470
+ },
2471
+ "environments": [
2472
+ "stylized_city_night",
2473
+ "stylized_city_sunset"
2474
+ ]
2475
+ },
2476
+ {
2477
+ "file": "syn/train/data-00218-of-00451.parquet",
2478
+ "rows": 1000,
2479
+ "bytes": 1930727965,
2480
+ "classes": {
2481
+ "crosswalk": 1000
2482
+ },
2483
+ "environments": [
2484
+ "stylized_city_sunset"
2485
+ ]
2486
+ },
2487
+ {
2488
+ "file": "syn/train/data-00219-of-00451.parquet",
2489
+ "rows": 1000,
2490
+ "bytes": 1784316682,
2491
+ "classes": {
2492
+ "crosswalk": 1000
2493
+ },
2494
+ "environments": [
2495
+ "stylized_city_sunset"
2496
+ ]
2497
+ },
2498
+ {
2499
+ "file": "syn/train/data-00220-of-00451.parquet",
2500
+ "rows": 1000,
2501
+ "bytes": 1802621620,
2502
+ "classes": {
2503
+ "crosswalk": 1000
2504
+ },
2505
+ "environments": [
2506
+ "stylized_city_sunset"
2507
+ ]
2508
+ },
2509
+ {
2510
+ "file": "syn/train/data-00221-of-00451.parquet",
2511
+ "rows": 1000,
2512
+ "bytes": 1744788949,
2513
+ "classes": {
2514
+ "crosswalk": 1000
2515
+ },
2516
+ "environments": [
2517
+ "stylized_city_sunset",
2518
+ "suburbs"
2519
+ ]
2520
+ },
2521
+ {
2522
+ "file": "syn/train/data-00222-of-00451.parquet",
2523
+ "rows": 1000,
2524
+ "bytes": 1567411216,
2525
+ "classes": {
2526
+ "crosswalk": 1000
2527
+ },
2528
+ "environments": [
2529
+ "suburbs"
2530
+ ]
2531
+ },
2532
+ {
2533
+ "file": "syn/train/data-00223-of-00451.parquet",
2534
+ "rows": 1000,
2535
+ "bytes": 1416392965,
2536
+ "classes": {
2537
+ "crosswalk": 1000
2538
+ },
2539
+ "environments": [
2540
+ "suburbs",
2541
+ "suburbs_night"
2542
+ ]
2543
+ },
2544
+ {
2545
+ "file": "syn/train/data-00224-of-00451.parquet",
2546
+ "rows": 1000,
2547
+ "bytes": 1410436963,
2548
+ "classes": {
2549
+ "crosswalk": 1000
2550
+ },
2551
+ "environments": [
2552
+ "suburbs_night"
2553
+ ]
2554
+ },
2555
+ {
2556
+ "file": "syn/train/data-00225-of-00451.parquet",
2557
+ "rows": 1000,
2558
+ "bytes": 1344494333,
2559
+ "classes": {
2560
+ "crosswalk": 1000
2561
+ },
2562
+ "environments": [
2563
+ "suburbs_night"
2564
+ ]
2565
+ },
2566
+ {
2567
+ "file": "syn/train/data-00226-of-00451.parquet",
2568
+ "rows": 1000,
2569
+ "bytes": 1493967799,
2570
+ "classes": {
2571
+ "crosswalk": 1000
2572
+ },
2573
+ "environments": [
2574
+ "suburbs_night",
2575
+ "suburbs_sunset"
2576
+ ]
2577
+ },
2578
+ {
2579
+ "file": "syn/train/data-00227-of-00451.parquet",
2580
+ "rows": 1000,
2581
+ "bytes": 1538487308,
2582
+ "classes": {
2583
+ "crosswalk": 1000
2584
+ },
2585
+ "environments": [
2586
+ "suburbs_sunset"
2587
+ ]
2588
+ },
2589
+ {
2590
+ "file": "syn/train/data-00228-of-00451.parquet",
2591
+ "rows": 1000,
2592
+ "bytes": 1462522081,
2593
+ "classes": {
2594
+ "crosswalk": 1000
2595
+ },
2596
+ "environments": [
2597
+ "suburbs_sunset",
2598
+ "tokyo_street"
2599
+ ]
2600
+ },
2601
+ {
2602
+ "file": "syn/train/data-00229-of-00451.parquet",
2603
+ "rows": 1000,
2604
+ "bytes": 2184109139,
2605
+ "classes": {
2606
+ "crosswalk": 1000
2607
+ },
2608
+ "environments": [
2609
+ "tokyo_street"
2610
+ ]
2611
+ },
2612
+ {
2613
+ "file": "syn/train/data-00230-of-00451.parquet",
2614
+ "rows": 1000,
2615
+ "bytes": 2103176217,
2616
+ "classes": {
2617
+ "crosswalk": 1000
2618
+ },
2619
+ "environments": [
2620
+ "tokyo_street",
2621
+ "tokyo_street_sunset"
2622
+ ]
2623
+ },
2624
+ {
2625
+ "file": "syn/train/data-00231-of-00451.parquet",
2626
+ "rows": 1000,
2627
+ "bytes": 2207665707,
2628
+ "classes": {
2629
+ "crosswalk": 1000
2630
+ },
2631
+ "environments": [
2632
+ "tokyo_street_sunset"
2633
+ ]
2634
+ },
2635
+ {
2636
+ "file": "syn/train/data-00232-of-00451.parquet",
2637
+ "rows": 1000,
2638
+ "bytes": 2144519289,
2639
+ "classes": {
2640
+ "crosswalk": 1000
2641
+ },
2642
+ "environments": [
2643
+ "tokyo_street_sunset"
2644
+ ]
2645
+ },
2646
+ {
2647
+ "file": "syn/train/data-00233-of-00451.parquet",
2648
+ "rows": 1000,
2649
+ "bytes": 1849929740,
2650
+ "classes": {
2651
+ "crosswalk": 1000
2652
+ },
2653
+ "environments": [
2654
+ "tokyo_street_sunset",
2655
+ "village"
2656
+ ]
2657
+ },
2658
+ {
2659
+ "file": "syn/train/data-00234-of-00451.parquet",
2660
+ "rows": 1000,
2661
+ "bytes": 1945582167,
2662
+ "classes": {
2663
+ "crosswalk": 1000
2664
+ },
2665
+ "environments": [
2666
+ "village"
2667
+ ]
2668
+ },
2669
+ {
2670
+ "file": "syn/train/data-00235-of-00451.parquet",
2671
+ "rows": 1000,
2672
+ "bytes": 1956996146,
2673
+ "classes": {
2674
+ "crosswalk": 1000
2675
+ },
2676
+ "environments": [
2677
+ "village",
2678
+ "village_sunset"
2679
+ ]
2680
+ },
2681
+ {
2682
+ "file": "syn/train/data-00236-of-00451.parquet",
2683
+ "rows": 1000,
2684
+ "bytes": 1976196186,
2685
+ "classes": {
2686
+ "crosswalk": 1000
2687
+ },
2688
+ "environments": [
2689
+ "village_sunset"
2690
+ ]
2691
+ },
2692
+ {
2693
+ "file": "syn/train/data-00237-of-00451.parquet",
2694
+ "rows": 1000,
2695
+ "bytes": 2016692905,
2696
+ "classes": {
2697
+ "crosswalk": 880,
2698
+ "door_button": 120
2699
+ },
2700
+ "environments": [
2701
+ "downtown_west",
2702
+ "village_sunset"
2703
+ ]
2704
+ },
2705
+ {
2706
+ "file": "syn/train/data-00238-of-00451.parquet",
2707
+ "rows": 1000,
2708
+ "bytes": 1635865206,
2709
+ "classes": {
2710
+ "door_button": 1000
2711
+ },
2712
+ "environments": [
2713
+ "downtown_west"
2714
+ ]
2715
+ },
2716
+ {
2717
+ "file": "syn/train/data-00239-of-00451.parquet",
2718
+ "rows": 1000,
2719
+ "bytes": 1535730771,
2720
+ "classes": {
2721
+ "door_button": 1000
2722
+ },
2723
+ "environments": [
2724
+ "downtown_west"
2725
+ ]
2726
+ },
2727
+ {
2728
+ "file": "syn/train/data-00240-of-00451.parquet",
2729
+ "rows": 1000,
2730
+ "bytes": 1448534426,
2731
+ "classes": {
2732
+ "door_button": 1000
2733
+ },
2734
+ "environments": [
2735
+ "downtown_west"
2736
+ ]
2737
+ },
2738
+ {
2739
+ "file": "syn/train/data-00241-of-00451.parquet",
2740
+ "rows": 1000,
2741
+ "bytes": 1603486931,
2742
+ "classes": {
2743
+ "door_button": 1000
2744
+ },
2745
+ "environments": [
2746
+ "downtown_west"
2747
+ ]
2748
+ },
2749
+ {
2750
+ "file": "syn/train/data-00242-of-00451.parquet",
2751
+ "rows": 1000,
2752
+ "bytes": 1558840128,
2753
+ "classes": {
2754
+ "door_button": 1000
2755
+ },
2756
+ "environments": [
2757
+ "downtown_west",
2758
+ "downtown_west_night"
2759
+ ]
2760
+ },
2761
+ {
2762
+ "file": "syn/train/data-00243-of-00451.parquet",
2763
+ "rows": 1000,
2764
+ "bytes": 1369179862,
2765
+ "classes": {
2766
+ "door_button": 1000
2767
+ },
2768
+ "environments": [
2769
+ "downtown_west_night"
2770
+ ]
2771
+ },
2772
+ {
2773
+ "file": "syn/train/data-00244-of-00451.parquet",
2774
+ "rows": 1000,
2775
+ "bytes": 1300276572,
2776
+ "classes": {
2777
+ "door_button": 1000
2778
+ },
2779
+ "environments": [
2780
+ "downtown_west_night"
2781
+ ]
2782
+ },
2783
+ {
2784
+ "file": "syn/train/data-00245-of-00451.parquet",
2785
+ "rows": 1000,
2786
+ "bytes": 1228799845,
2787
+ "classes": {
2788
+ "door_button": 1000
2789
+ },
2790
+ "environments": [
2791
+ "downtown_west_night"
2792
+ ]
2793
+ },
2794
+ {
2795
+ "file": "syn/train/data-00246-of-00451.parquet",
2796
+ "rows": 1000,
2797
+ "bytes": 1313208733,
2798
+ "classes": {
2799
+ "door_button": 1000
2800
+ },
2801
+ "environments": [
2802
+ "downtown_west_night"
2803
+ ]
2804
+ },
2805
+ {
2806
+ "file": "syn/train/data-00247-of-00451.parquet",
2807
+ "rows": 1000,
2808
+ "bytes": 1524990268,
2809
+ "classes": {
2810
+ "door_button": 1000
2811
+ },
2812
+ "environments": [
2813
+ "downtown_west_night",
2814
+ "downtown_west_sunset"
2815
+ ]
2816
+ },
2817
+ {
2818
+ "file": "syn/train/data-00248-of-00451.parquet",
2819
+ "rows": 1000,
2820
+ "bytes": 1490411232,
2821
+ "classes": {
2822
+ "door_button": 1000
2823
+ },
2824
+ "environments": [
2825
+ "downtown_west_sunset"
2826
+ ]
2827
+ },
2828
+ {
2829
+ "file": "syn/train/data-00249-of-00451.parquet",
2830
+ "rows": 1000,
2831
+ "bytes": 1389261944,
2832
+ "classes": {
2833
+ "door_button": 1000
2834
+ },
2835
+ "environments": [
2836
+ "downtown_west_sunset"
2837
+ ]
2838
+ },
2839
+ {
2840
+ "file": "syn/train/data-00250-of-00451.parquet",
2841
+ "rows": 1000,
2842
+ "bytes": 1497386718,
2843
+ "classes": {
2844
+ "door_button": 1000
2845
+ },
2846
+ "environments": [
2847
+ "downtown_west_sunset"
2848
+ ]
2849
+ },
2850
+ {
2851
+ "file": "syn/train/data-00251-of-00451.parquet",
2852
+ "rows": 1000,
2853
+ "bytes": 1583086341,
2854
+ "classes": {
2855
+ "door_button": 1000
2856
+ },
2857
+ "environments": [
2858
+ "downtown_west_sunset"
2859
+ ]
2860
+ },
2861
+ {
2862
+ "file": "syn/train/data-00252-of-00451.parquet",
2863
+ "rows": 1000,
2864
+ "bytes": 1728131259,
2865
+ "classes": {
2866
+ "door_button": 1000
2867
+ },
2868
+ "environments": [
2869
+ "downtown_west_sunset",
2870
+ "hong_kong_street"
2871
+ ]
2872
+ },
2873
+ {
2874
+ "file": "syn/train/data-00253-of-00451.parquet",
2875
+ "rows": 1000,
2876
+ "bytes": 1665022358,
2877
+ "classes": {
2878
+ "door_button": 1000
2879
+ },
2880
+ "environments": [
2881
+ "hong_kong_street"
2882
+ ]
2883
+ },
2884
+ {
2885
+ "file": "syn/train/data-00254-of-00451.parquet",
2886
+ "rows": 1000,
2887
+ "bytes": 1656549997,
2888
+ "classes": {
2889
+ "door_button": 1000
2890
+ },
2891
+ "environments": [
2892
+ "hong_kong_street"
2893
+ ]
2894
+ },
2895
+ {
2896
+ "file": "syn/train/data-00255-of-00451.parquet",
2897
+ "rows": 1000,
2898
+ "bytes": 1722626153,
2899
+ "classes": {
2900
+ "door_button": 1000
2901
+ },
2902
+ "environments": [
2903
+ "hong_kong_street"
2904
+ ]
2905
+ },
2906
+ {
2907
+ "file": "syn/train/data-00256-of-00451.parquet",
2908
+ "rows": 1000,
2909
+ "bytes": 1626749841,
2910
+ "classes": {
2911
+ "door_button": 1000
2912
+ },
2913
+ "environments": [
2914
+ "hong_kong_street"
2915
+ ]
2916
+ },
2917
+ {
2918
+ "file": "syn/train/data-00257-of-00451.parquet",
2919
+ "rows": 1000,
2920
+ "bytes": 1615882785,
2921
+ "classes": {
2922
+ "door_button": 1000
2923
+ },
2924
+ "environments": [
2925
+ "hong_kong_street",
2926
+ "hong_kong_street_sunset"
2927
+ ]
2928
+ },
2929
+ {
2930
+ "file": "syn/train/data-00258-of-00451.parquet",
2931
+ "rows": 1000,
2932
+ "bytes": 1619513063,
2933
+ "classes": {
2934
+ "door_button": 1000
2935
+ },
2936
+ "environments": [
2937
+ "hong_kong_street_sunset"
2938
+ ]
2939
+ },
2940
+ {
2941
+ "file": "syn/train/data-00259-of-00451.parquet",
2942
+ "rows": 1000,
2943
+ "bytes": 1530512842,
2944
+ "classes": {
2945
+ "door_button": 1000
2946
+ },
2947
+ "environments": [
2948
+ "hong_kong_street_sunset"
2949
+ ]
2950
+ },
2951
+ {
2952
+ "file": "syn/train/data-00260-of-00451.parquet",
2953
+ "rows": 1000,
2954
+ "bytes": 1632878142,
2955
+ "classes": {
2956
+ "door_button": 1000
2957
+ },
2958
+ "environments": [
2959
+ "hong_kong_street_sunset"
2960
+ ]
2961
+ },
2962
+ {
2963
+ "file": "syn/train/data-00261-of-00451.parquet",
2964
+ "rows": 1000,
2965
+ "bytes": 1697937668,
2966
+ "classes": {
2967
+ "door_button": 1000
2968
+ },
2969
+ "environments": [
2970
+ "hong_kong_street_sunset"
2971
+ ]
2972
+ },
2973
+ {
2974
+ "file": "syn/train/data-00262-of-00451.parquet",
2975
+ "rows": 1000,
2976
+ "bytes": 1458583007,
2977
+ "classes": {
2978
+ "door_button": 1000
2979
+ },
2980
+ "environments": [
2981
+ "hong_kong_street_sunset",
2982
+ "skyscraper"
2983
+ ]
2984
+ },
2985
+ {
2986
+ "file": "syn/train/data-00263-of-00451.parquet",
2987
+ "rows": 1000,
2988
+ "bytes": 1571019832,
2989
+ "classes": {
2990
+ "door_button": 1000
2991
+ },
2992
+ "environments": [
2993
+ "skyscraper"
2994
+ ]
2995
+ },
2996
+ {
2997
+ "file": "syn/train/data-00264-of-00451.parquet",
2998
+ "rows": 1000,
2999
+ "bytes": 1583225709,
3000
+ "classes": {
3001
+ "door_button": 1000
3002
+ },
3003
+ "environments": [
3004
+ "skyscraper"
3005
+ ]
3006
+ },
3007
+ {
3008
+ "file": "syn/train/data-00265-of-00451.parquet",
3009
+ "rows": 1000,
3010
+ "bytes": 1482673452,
3011
+ "classes": {
3012
+ "door_button": 1000
3013
+ },
3014
+ "environments": [
3015
+ "skyscraper"
3016
+ ]
3017
+ },
3018
+ {
3019
+ "file": "syn/train/data-00266-of-00451.parquet",
3020
+ "rows": 1000,
3021
+ "bytes": 1461721468,
3022
+ "classes": {
3023
+ "door_button": 1000
3024
+ },
3025
+ "environments": [
3026
+ "skyscraper"
3027
+ ]
3028
+ },
3029
+ {
3030
+ "file": "syn/train/data-00267-of-00451.parquet",
3031
+ "rows": 1000,
3032
+ "bytes": 1741392114,
3033
+ "classes": {
3034
+ "door_button": 1000
3035
+ },
3036
+ "environments": [
3037
+ "skyscraper",
3038
+ "small_city"
3039
+ ]
3040
+ },
3041
+ {
3042
+ "file": "syn/train/data-00268-of-00451.parquet",
3043
+ "rows": 1000,
3044
+ "bytes": 1743679822,
3045
+ "classes": {
3046
+ "door_button": 1000
3047
+ },
3048
+ "environments": [
3049
+ "small_city"
3050
+ ]
3051
+ },
3052
+ {
3053
+ "file": "syn/train/data-00269-of-00451.parquet",
3054
+ "rows": 1000,
3055
+ "bytes": 1758179250,
3056
+ "classes": {
3057
+ "door_button": 1000
3058
+ },
3059
+ "environments": [
3060
+ "small_city"
3061
+ ]
3062
+ },
3063
+ {
3064
+ "file": "syn/train/data-00270-of-00451.parquet",
3065
+ "rows": 1000,
3066
+ "bytes": 1753831228,
3067
+ "classes": {
3068
+ "door_button": 1000
3069
+ },
3070
+ "environments": [
3071
+ "small_city",
3072
+ "small_city_sunset"
3073
+ ]
3074
+ },
3075
+ {
3076
+ "file": "syn/train/data-00271-of-00451.parquet",
3077
+ "rows": 1000,
3078
+ "bytes": 1749263105,
3079
+ "classes": {
3080
+ "door_button": 1000
3081
+ },
3082
+ "environments": [
3083
+ "small_city_sunset"
3084
+ ]
3085
+ },
3086
+ {
3087
+ "file": "syn/train/data-00272-of-00451.parquet",
3088
+ "rows": 1000,
3089
+ "bytes": 1696265064,
3090
+ "classes": {
3091
+ "door_button": 1000
3092
+ },
3093
+ "environments": [
3094
+ "small_city_sunset"
3095
+ ]
3096
+ },
3097
+ {
3098
+ "file": "syn/train/data-00273-of-00451.parquet",
3099
+ "rows": 1000,
3100
+ "bytes": 1545983979,
3101
+ "classes": {
3102
+ "door_button": 1000
3103
+ },
3104
+ "environments": [
3105
+ "small_city_sunset"
3106
+ ]
3107
+ },
3108
+ {
3109
+ "file": "syn/train/data-00274-of-00451.parquet",
3110
+ "rows": 1000,
3111
+ "bytes": 1658127820,
3112
+ "classes": {
3113
+ "door_button": 1000
3114
+ },
3115
+ "environments": [
3116
+ "small_city_sunset",
3117
+ "tokyo_street"
3118
+ ]
3119
+ },
3120
+ {
3121
+ "file": "syn/train/data-00275-of-00451.parquet",
3122
+ "rows": 1000,
3123
+ "bytes": 1490830839,
3124
+ "classes": {
3125
+ "door_button": 1000
3126
+ },
3127
+ "environments": [
3128
+ "tokyo_street"
3129
+ ]
3130
+ },
3131
+ {
3132
+ "file": "syn/train/data-00276-of-00451.parquet",
3133
+ "rows": 1000,
3134
+ "bytes": 1497974867,
3135
+ "classes": {
3136
+ "door_button": 1000
3137
+ },
3138
+ "environments": [
3139
+ "tokyo_street"
3140
+ ]
3141
+ },
3142
+ {
3143
+ "file": "syn/train/data-00277-of-00451.parquet",
3144
+ "rows": 1000,
3145
+ "bytes": 1528474286,
3146
+ "classes": {
3147
+ "door_button": 1000
3148
+ },
3149
+ "environments": [
3150
+ "tokyo_street"
3151
+ ]
3152
+ },
3153
+ {
3154
+ "file": "syn/train/data-00278-of-00451.parquet",
3155
+ "rows": 1000,
3156
+ "bytes": 1564838121,
3157
+ "classes": {
3158
+ "door_button": 1000
3159
+ },
3160
+ "environments": [
3161
+ "tokyo_street",
3162
+ "tokyo_street_sunset"
3163
+ ]
3164
+ },
3165
+ {
3166
+ "file": "syn/train/data-00279-of-00451.parquet",
3167
+ "rows": 1000,
3168
+ "bytes": 1521723863,
3169
+ "classes": {
3170
+ "door_button": 1000
3171
+ },
3172
+ "environments": [
3173
+ "tokyo_street_sunset"
3174
+ ]
3175
+ },
3176
+ {
3177
+ "file": "syn/train/data-00280-of-00451.parquet",
3178
+ "rows": 1000,
3179
+ "bytes": 1478954483,
3180
+ "classes": {
3181
+ "door_button": 1000
3182
+ },
3183
+ "environments": [
3184
+ "tokyo_street_sunset"
3185
+ ]
3186
+ },
3187
+ {
3188
+ "file": "syn/train/data-00281-of-00451.parquet",
3189
+ "rows": 1000,
3190
+ "bytes": 1513113587,
3191
+ "classes": {
3192
+ "door_button": 1000
3193
+ },
3194
+ "environments": [
3195
+ "tokyo_street_sunset"
3196
+ ]
3197
+ },
3198
+ {
3199
+ "file": "syn/train/data-00282-of-00451.parquet",
3200
+ "rows": 1000,
3201
+ "bytes": 1559311676,
3202
+ "classes": {
3203
+ "door_button": 1000
3204
+ },
3205
+ "environments": [
3206
+ "tokyo_street_sunset"
3207
+ ]
3208
+ },
3209
+ {
3210
+ "file": "syn/train/data-00283-of-00451.parquet",
3211
+ "rows": 1000,
3212
+ "bytes": 1412475418,
3213
+ "classes": {
3214
+ "door_button": 240,
3215
+ "elevator": 760
3216
+ },
3217
+ "environments": [
3218
+ "airport",
3219
+ "tokyo_street_sunset"
3220
+ ]
3221
+ },
3222
+ {
3223
+ "file": "syn/train/data-00284-of-00451.parquet",
3224
+ "rows": 1000,
3225
+ "bytes": 1392063548,
3226
+ "classes": {
3227
+ "elevator": 1000
3228
+ },
3229
+ "environments": [
3230
+ "airport"
3231
+ ]
3232
+ },
3233
+ {
3234
+ "file": "syn/train/data-00285-of-00451.parquet",
3235
+ "rows": 1000,
3236
+ "bytes": 980998569,
3237
+ "classes": {
3238
+ "elevator": 1000
3239
+ },
3240
+ "environments": [
3241
+ "airport",
3242
+ "elevator_module"
3243
+ ]
3244
+ },
3245
+ {
3246
+ "file": "syn/train/data-00286-of-00451.parquet",
3247
+ "rows": 1000,
3248
+ "bytes": 1244579876,
3249
+ "classes": {
3250
+ "elevator": 1000
3251
+ },
3252
+ "environments": [
3253
+ "elevator_module",
3254
+ "havana_street",
3255
+ "hospital_hallway"
3256
+ ]
3257
+ },
3258
+ {
3259
+ "file": "syn/train/data-00287-of-00451.parquet",
3260
+ "rows": 1000,
3261
+ "bytes": 170016048,
3262
+ "classes": {
3263
+ "elevator": 1000
3264
+ },
3265
+ "environments": [
3266
+ "hospital_hallway"
3267
+ ]
3268
+ },
3269
+ {
3270
+ "file": "syn/train/data-00288-of-00451.parquet",
3271
+ "rows": 1000,
3272
+ "bytes": 161801728,
3273
+ "classes": {
3274
+ "elevator": 1000
3275
+ },
3276
+ "environments": [
3277
+ "hospital_hallway"
3278
+ ]
3279
+ },
3280
+ {
3281
+ "file": "syn/train/data-00289-of-00451.parquet",
3282
+ "rows": 1000,
3283
+ "bytes": 1203358662,
3284
+ "classes": {
3285
+ "elevator": 1000
3286
+ },
3287
+ "environments": [
3288
+ "hospital_hallway",
3289
+ "hotel_corridor"
3290
+ ]
3291
+ },
3292
+ {
3293
+ "file": "syn/train/data-00290-of-00451.parquet",
3294
+ "rows": 1000,
3295
+ "bytes": 469510529,
3296
+ "classes": {
3297
+ "elevator": 1000
3298
+ },
3299
+ "environments": [
3300
+ "hotel_corridor",
3301
+ "hotel_reception",
3302
+ "lobby"
3303
+ ]
3304
+ },
3305
+ {
3306
+ "file": "syn/train/data-00291-of-00451.parquet",
3307
+ "rows": 1000,
3308
+ "bytes": 1218589399,
3309
+ "classes": {
3310
+ "elevator": 1000
3311
+ },
3312
+ "environments": [
3313
+ "lobby"
3314
+ ]
3315
+ },
3316
+ {
3317
+ "file": "syn/train/data-00292-of-00451.parquet",
3318
+ "rows": 1000,
3319
+ "bytes": 1518840031,
3320
+ "classes": {
3321
+ "elevator": 1000
3322
+ },
3323
+ "environments": [
3324
+ "lobby",
3325
+ "miami_beach"
3326
+ ]
3327
+ },
3328
+ {
3329
+ "file": "syn/train/data-00293-of-00451.parquet",
3330
+ "rows": 1000,
3331
+ "bytes": 1668372642,
3332
+ "classes": {
3333
+ "elevator": 1000
3334
+ },
3335
+ "environments": [
3336
+ "miami_beach",
3337
+ "shopping_mall"
3338
+ ]
3339
+ },
3340
+ {
3341
+ "file": "syn/train/data-00294-of-00451.parquet",
3342
+ "rows": 1000,
3343
+ "bytes": 1718322038,
3344
+ "classes": {
3345
+ "elevator": 1000
3346
+ },
3347
+ "environments": [
3348
+ "shopping_mall"
3349
+ ]
3350
+ },
3351
+ {
3352
+ "file": "syn/train/data-00295-of-00451.parquet",
3353
+ "rows": 1000,
3354
+ "bytes": 1593835472,
3355
+ "classes": {
3356
+ "elevator": 1000
3357
+ },
3358
+ "environments": [
3359
+ "shopping_mall"
3360
+ ]
3361
+ },
3362
+ {
3363
+ "file": "syn/train/data-00296-of-00451.parquet",
3364
+ "rows": 1000,
3365
+ "bytes": 1316652864,
3366
+ "classes": {
3367
+ "elevator": 1000
3368
+ },
3369
+ "environments": [
3370
+ "shopping_mall",
3371
+ "skyscraper"
3372
+ ]
3373
+ },
3374
+ {
3375
+ "file": "syn/train/data-00297-of-00451.parquet",
3376
+ "rows": 1000,
3377
+ "bytes": 1404174393,
3378
+ "classes": {
3379
+ "elevator": 1000
3380
+ },
3381
+ "environments": [
3382
+ "skyscraper",
3383
+ "station_seoul"
3384
+ ]
3385
+ },
3386
+ {
3387
+ "file": "syn/train/data-00298-of-00451.parquet",
3388
+ "rows": 1000,
3389
+ "bytes": 1504933464,
3390
+ "classes": {
3391
+ "elevator": 1000
3392
+ },
3393
+ "environments": [
3394
+ "station_seoul"
3395
+ ]
3396
+ },
3397
+ {
3398
+ "file": "syn/train/data-00299-of-00451.parquet",
3399
+ "rows": 1000,
3400
+ "bytes": 1161058243,
3401
+ "classes": {
3402
+ "elevator": 1000
3403
+ },
3404
+ "environments": [
3405
+ "station_seoul"
3406
+ ]
3407
+ },
3408
+ {
3409
+ "file": "syn/train/data-00300-of-00451.parquet",
3410
+ "rows": 1000,
3411
+ "bytes": 1276804690,
3412
+ "classes": {
3413
+ "elevator": 1000
3414
+ },
3415
+ "environments": [
3416
+ "station_seoul",
3417
+ "subway_station"
3418
+ ]
3419
+ },
3420
+ {
3421
+ "file": "syn/train/data-00301-of-00451.parquet",
3422
+ "rows": 1000,
3423
+ "bytes": 1248999762,
3424
+ "classes": {
3425
+ "elevator": 1000
3426
+ },
3427
+ "environments": [
3428
+ "subway_station"
3429
+ ]
3430
+ },
3431
+ {
3432
+ "file": "syn/train/data-00302-of-00451.parquet",
3433
+ "rows": 1000,
3434
+ "bytes": 1298220084,
3435
+ "classes": {
3436
+ "elevator": 1000
3437
+ },
3438
+ "environments": [
3439
+ "subway_station"
3440
+ ]
3441
+ },
3442
+ {
3443
+ "file": "syn/train/data-00303-of-00451.parquet",
3444
+ "rows": 1000,
3445
+ "bytes": 1313942546,
3446
+ "classes": {
3447
+ "elevator": 1000
3448
+ },
3449
+ "environments": [
3450
+ "subway_station"
3451
+ ]
3452
+ },
3453
+ {
3454
+ "file": "syn/train/data-00304-of-00451.parquet",
3455
+ "rows": 1000,
3456
+ "bytes": 1510282976,
3457
+ "classes": {
3458
+ "elevator": 1000
3459
+ },
3460
+ "environments": [
3461
+ "subway_station",
3462
+ "tokyo_street",
3463
+ "train_terminal"
3464
+ ]
3465
+ },
3466
+ {
3467
+ "file": "syn/train/data-00305-of-00451.parquet",
3468
+ "rows": 1000,
3469
+ "bytes": 1200919061,
3470
+ "classes": {
3471
+ "elevator": 1000
3472
+ },
3473
+ "environments": [
3474
+ "train_terminal"
3475
+ ]
3476
+ },
3477
+ {
3478
+ "file": "syn/train/data-00306-of-00451.parquet",
3479
+ "rows": 1000,
3480
+ "bytes": 1339856056,
3481
+ "classes": {
3482
+ "elevator": 590,
3483
+ "elevator_button": 410
3484
+ },
3485
+ "environments": [
3486
+ "airport",
3487
+ "train_terminal"
3488
+ ]
3489
+ },
3490
+ {
3491
+ "file": "syn/train/data-00307-of-00451.parquet",
3492
+ "rows": 1000,
3493
+ "bytes": 1396534514,
3494
+ "classes": {
3495
+ "elevator_button": 1000
3496
+ },
3497
+ "environments": [
3498
+ "airport"
3499
+ ]
3500
+ },
3501
+ {
3502
+ "file": "syn/train/data-00308-of-00451.parquet",
3503
+ "rows": 1000,
3504
+ "bytes": 1227423361,
3505
+ "classes": {
3506
+ "elevator_button": 1000
3507
+ },
3508
+ "environments": [
3509
+ "airport",
3510
+ "elevator_module"
3511
+ ]
3512
+ },
3513
+ {
3514
+ "file": "syn/train/data-00309-of-00451.parquet",
3515
+ "rows": 1000,
3516
+ "bytes": 1057916385,
3517
+ "classes": {
3518
+ "elevator_button": 1000
3519
+ },
3520
+ "environments": [
3521
+ "elevator_module",
3522
+ "havana_street"
3523
+ ]
3524
+ },
3525
+ {
3526
+ "file": "syn/train/data-00310-of-00451.parquet",
3527
+ "rows": 1000,
3528
+ "bytes": 511143797,
3529
+ "classes": {
3530
+ "elevator_button": 1000
3531
+ },
3532
+ "environments": [
3533
+ "havana_street",
3534
+ "hospital_hallway"
3535
+ ]
3536
+ },
3537
+ {
3538
+ "file": "syn/train/data-00311-of-00451.parquet",
3539
+ "rows": 1000,
3540
+ "bytes": 167930588,
3541
+ "classes": {
3542
+ "elevator_button": 1000
3543
+ },
3544
+ "environments": [
3545
+ "hospital_hallway"
3546
+ ]
3547
+ },
3548
+ {
3549
+ "file": "syn/train/data-00312-of-00451.parquet",
3550
+ "rows": 1000,
3551
+ "bytes": 806193038,
3552
+ "classes": {
3553
+ "elevator_button": 1000
3554
+ },
3555
+ "environments": [
3556
+ "hospital_hallway",
3557
+ "hotel_corridor"
3558
+ ]
3559
+ },
3560
+ {
3561
+ "file": "syn/train/data-00313-of-00451.parquet",
3562
+ "rows": 1000,
3563
+ "bytes": 637408570,
3564
+ "classes": {
3565
+ "elevator_button": 1000
3566
+ },
3567
+ "environments": [
3568
+ "hotel_corridor",
3569
+ "hotel_reception"
3570
+ ]
3571
+ },
3572
+ {
3573
+ "file": "syn/train/data-00314-of-00451.parquet",
3574
+ "rows": 1000,
3575
+ "bytes": 944545305,
3576
+ "classes": {
3577
+ "elevator_button": 1000
3578
+ },
3579
+ "environments": [
3580
+ "hotel_reception",
3581
+ "lobby"
3582
+ ]
3583
+ },
3584
+ {
3585
+ "file": "syn/train/data-00315-of-00451.parquet",
3586
+ "rows": 1000,
3587
+ "bytes": 1405965599,
3588
+ "classes": {
3589
+ "elevator_button": 1000
3590
+ },
3591
+ "environments": [
3592
+ "lobby",
3593
+ "miami_beach"
3594
+ ]
3595
+ },
3596
+ {
3597
+ "file": "syn/train/data-00316-of-00451.parquet",
3598
+ "rows": 1000,
3599
+ "bytes": 1600371403,
3600
+ "classes": {
3601
+ "elevator_button": 1000
3602
+ },
3603
+ "environments": [
3604
+ "miami_beach",
3605
+ "shopping_mall"
3606
+ ]
3607
+ },
3608
+ {
3609
+ "file": "syn/train/data-00317-of-00451.parquet",
3610
+ "rows": 1000,
3611
+ "bytes": 1715094995,
3612
+ "classes": {
3613
+ "elevator_button": 1000
3614
+ },
3615
+ "environments": [
3616
+ "shopping_mall"
3617
+ ]
3618
+ },
3619
+ {
3620
+ "file": "syn/train/data-00318-of-00451.parquet",
3621
+ "rows": 1000,
3622
+ "bytes": 1679927644,
3623
+ "classes": {
3624
+ "elevator_button": 1000
3625
+ },
3626
+ "environments": [
3627
+ "shopping_mall"
3628
+ ]
3629
+ },
3630
+ {
3631
+ "file": "syn/train/data-00319-of-00451.parquet",
3632
+ "rows": 1000,
3633
+ "bytes": 1367870484,
3634
+ "classes": {
3635
+ "elevator_button": 1000
3636
+ },
3637
+ "environments": [
3638
+ "shopping_mall",
3639
+ "skyscraper"
3640
+ ]
3641
+ },
3642
+ {
3643
+ "file": "syn/train/data-00320-of-00451.parquet",
3644
+ "rows": 1000,
3645
+ "bytes": 1324471551,
3646
+ "classes": {
3647
+ "elevator_button": 1000
3648
+ },
3649
+ "environments": [
3650
+ "skyscraper"
3651
+ ]
3652
+ },
3653
+ {
3654
+ "file": "syn/train/data-00321-of-00451.parquet",
3655
+ "rows": 1000,
3656
+ "bytes": 1579283898,
3657
+ "classes": {
3658
+ "elevator_button": 1000
3659
+ },
3660
+ "environments": [
3661
+ "skyscraper",
3662
+ "station_seoul"
3663
+ ]
3664
+ },
3665
+ {
3666
+ "file": "syn/train/data-00322-of-00451.parquet",
3667
+ "rows": 1000,
3668
+ "bytes": 1084003504,
3669
+ "classes": {
3670
+ "elevator_button": 1000
3671
+ },
3672
+ "environments": [
3673
+ "station_seoul"
3674
+ ]
3675
+ },
3676
+ {
3677
+ "file": "syn/train/data-00323-of-00451.parquet",
3678
+ "rows": 1000,
3679
+ "bytes": 1309059864,
3680
+ "classes": {
3681
+ "elevator_button": 1000
3682
+ },
3683
+ "environments": [
3684
+ "station_seoul",
3685
+ "subway_station"
3686
+ ]
3687
+ },
3688
+ {
3689
+ "file": "syn/train/data-00324-of-00451.parquet",
3690
+ "rows": 1000,
3691
+ "bytes": 1211410961,
3692
+ "classes": {
3693
+ "elevator_button": 1000
3694
+ },
3695
+ "environments": [
3696
+ "subway_station"
3697
+ ]
3698
+ },
3699
+ {
3700
+ "file": "syn/train/data-00325-of-00451.parquet",
3701
+ "rows": 1000,
3702
+ "bytes": 1262146669,
3703
+ "classes": {
3704
+ "elevator_button": 1000
3705
+ },
3706
+ "environments": [
3707
+ "subway_station"
3708
+ ]
3709
+ },
3710
+ {
3711
+ "file": "syn/train/data-00326-of-00451.parquet",
3712
+ "rows": 1000,
3713
+ "bytes": 1307906825,
3714
+ "classes": {
3715
+ "elevator_button": 1000
3716
+ },
3717
+ "environments": [
3718
+ "subway_station"
3719
+ ]
3720
+ },
3721
+ {
3722
+ "file": "syn/train/data-00327-of-00451.parquet",
3723
+ "rows": 1000,
3724
+ "bytes": 1616956699,
3725
+ "classes": {
3726
+ "elevator_button": 1000
3727
+ },
3728
+ "environments": [
3729
+ "subway_station",
3730
+ "tokyo_street",
3731
+ "train_terminal"
3732
+ ]
3733
+ },
3734
+ {
3735
+ "file": "syn/train/data-00328-of-00451.parquet",
3736
+ "rows": 1000,
3737
+ "bytes": 1100666104,
3738
+ "classes": {
3739
+ "elevator_button": 1000
3740
+ },
3741
+ "environments": [
3742
+ "train_terminal"
3743
+ ]
3744
+ },
3745
+ {
3746
+ "file": "syn/train/data-00329-of-00451.parquet",
3747
+ "rows": 1000,
3748
+ "bytes": 1272790384,
3749
+ "classes": {
3750
+ "elevator_button": 940,
3751
+ "escalator": 60
3752
+ },
3753
+ "environments": [
3754
+ "airport",
3755
+ "train_terminal"
3756
+ ]
3757
+ },
3758
+ {
3759
+ "file": "syn/train/data-00330-of-00451.parquet",
3760
+ "rows": 1000,
3761
+ "bytes": 1475348155,
3762
+ "classes": {
3763
+ "escalator": 1000
3764
+ },
3765
+ "environments": [
3766
+ "airport"
3767
+ ]
3768
+ },
3769
+ {
3770
+ "file": "syn/train/data-00331-of-00451.parquet",
3771
+ "rows": 1000,
3772
+ "bytes": 1464964790,
3773
+ "classes": {
3774
+ "escalator": 1000
3775
+ },
3776
+ "environments": [
3777
+ "airport"
3778
+ ]
3779
+ },
3780
+ {
3781
+ "file": "syn/train/data-00332-of-00451.parquet",
3782
+ "rows": 1000,
3783
+ "bytes": 1509867425,
3784
+ "classes": {
3785
+ "escalator": 1000
3786
+ },
3787
+ "environments": [
3788
+ "airport",
3789
+ "shopping_mall"
3790
+ ]
3791
+ },
3792
+ {
3793
+ "file": "syn/train/data-00333-of-00451.parquet",
3794
+ "rows": 1000,
3795
+ "bytes": 1928752939,
3796
+ "classes": {
3797
+ "escalator": 1000
3798
+ },
3799
+ "environments": [
3800
+ "shopping_mall"
3801
+ ]
3802
+ },
3803
+ {
3804
+ "file": "syn/train/data-00334-of-00451.parquet",
3805
+ "rows": 1000,
3806
+ "bytes": 1955186882,
3807
+ "classes": {
3808
+ "escalator": 1000
3809
+ },
3810
+ "environments": [
3811
+ "shopping_mall"
3812
+ ]
3813
+ },
3814
+ {
3815
+ "file": "syn/train/data-00335-of-00451.parquet",
3816
+ "rows": 1000,
3817
+ "bytes": 1193640920,
3818
+ "classes": {
3819
+ "escalator": 1000
3820
+ },
3821
+ "environments": [
3822
+ "shopping_mall",
3823
+ "train_terminal"
3824
+ ]
3825
+ },
3826
+ {
3827
+ "file": "syn/train/data-00336-of-00451.parquet",
3828
+ "rows": 1000,
3829
+ "bytes": 1682405231,
3830
+ "classes": {
3831
+ "escalator": 360,
3832
+ "handrail": 640
3833
+ },
3834
+ "environments": [
3835
+ "downtown_west",
3836
+ "train_terminal"
3837
+ ]
3838
+ },
3839
+ {
3840
+ "file": "syn/train/data-00337-of-00451.parquet",
3841
+ "rows": 1000,
3842
+ "bytes": 1882058632,
3843
+ "classes": {
3844
+ "handrail": 1000
3845
+ },
3846
+ "environments": [
3847
+ "downtown_west"
3848
+ ]
3849
+ },
3850
+ {
3851
+ "file": "syn/train/data-00338-of-00451.parquet",
3852
+ "rows": 1000,
3853
+ "bytes": 1754890567,
3854
+ "classes": {
3855
+ "handrail": 1000
3856
+ },
3857
+ "environments": [
3858
+ "downtown_west"
3859
+ ]
3860
+ },
3861
+ {
3862
+ "file": "syn/train/data-00339-of-00451.parquet",
3863
+ "rows": 1000,
3864
+ "bytes": 1792462579,
3865
+ "classes": {
3866
+ "handrail": 1000
3867
+ },
3868
+ "environments": [
3869
+ "downtown_west",
3870
+ "downtown_west_night"
3871
+ ]
3872
+ },
3873
+ {
3874
+ "file": "syn/train/data-00340-of-00451.parquet",
3875
+ "rows": 1000,
3876
+ "bytes": 1626702494,
3877
+ "classes": {
3878
+ "handrail": 1000
3879
+ },
3880
+ "environments": [
3881
+ "downtown_west_night"
3882
+ ]
3883
+ },
3884
+ {
3885
+ "file": "syn/train/data-00341-of-00451.parquet",
3886
+ "rows": 1000,
3887
+ "bytes": 1456265948,
3888
+ "classes": {
3889
+ "handrail": 1000
3890
+ },
3891
+ "environments": [
3892
+ "downtown_west_night"
3893
+ ]
3894
+ },
3895
+ {
3896
+ "file": "syn/train/data-00342-of-00451.parquet",
3897
+ "rows": 1000,
3898
+ "bytes": 1645468518,
3899
+ "classes": {
3900
+ "handrail": 1000
3901
+ },
3902
+ "environments": [
3903
+ "downtown_west_night",
3904
+ "downtown_west_sunset"
3905
+ ]
3906
+ },
3907
+ {
3908
+ "file": "syn/train/data-00343-of-00451.parquet",
3909
+ "rows": 1000,
3910
+ "bytes": 1886502462,
3911
+ "classes": {
3912
+ "handrail": 1000
3913
+ },
3914
+ "environments": [
3915
+ "downtown_west_sunset"
3916
+ ]
3917
+ },
3918
+ {
3919
+ "file": "syn/train/data-00344-of-00451.parquet",
3920
+ "rows": 1000,
3921
+ "bytes": 1655619879,
3922
+ "classes": {
3923
+ "handrail": 1000
3924
+ },
3925
+ "environments": [
3926
+ "downtown_west_sunset"
3927
+ ]
3928
+ },
3929
+ {
3930
+ "file": "syn/train/data-00345-of-00451.parquet",
3931
+ "rows": 1000,
3932
+ "bytes": 1817309228,
3933
+ "classes": {
3934
+ "handrail": 1000
3935
+ },
3936
+ "environments": [
3937
+ "downtown_west_sunset",
3938
+ "hong_kong_street"
3939
+ ]
3940
+ },
3941
+ {
3942
+ "file": "syn/train/data-00346-of-00451.parquet",
3943
+ "rows": 1000,
3944
+ "bytes": 1639712611,
3945
+ "classes": {
3946
+ "handrail": 1000
3947
+ },
3948
+ "environments": [
3949
+ "hong_kong_street"
3950
+ ]
3951
+ },
3952
+ {
3953
+ "file": "syn/train/data-00347-of-00451.parquet",
3954
+ "rows": 1000,
3955
+ "bytes": 1602798945,
3956
+ "classes": {
3957
+ "handrail": 1000
3958
+ },
3959
+ "environments": [
3960
+ "hong_kong_street",
3961
+ "hong_kong_street_sunset"
3962
+ ]
3963
+ },
3964
+ {
3965
+ "file": "syn/train/data-00348-of-00451.parquet",
3966
+ "rows": 1000,
3967
+ "bytes": 1572812698,
3968
+ "classes": {
3969
+ "handrail": 1000
3970
+ },
3971
+ "environments": [
3972
+ "hong_kong_street_sunset"
3973
+ ]
3974
+ },
3975
+ {
3976
+ "file": "syn/train/data-00349-of-00451.parquet",
3977
+ "rows": 1000,
3978
+ "bytes": 1984929324,
3979
+ "classes": {
3980
+ "handrail": 1000
3981
+ },
3982
+ "environments": [
3983
+ "hong_kong_street_sunset",
3984
+ "miami_beach"
3985
+ ]
3986
+ },
3987
+ {
3988
+ "file": "syn/train/data-00350-of-00451.parquet",
3989
+ "rows": 1000,
3990
+ "bytes": 1937086331,
3991
+ "classes": {
3992
+ "handrail": 1000
3993
+ },
3994
+ "environments": [
3995
+ "miami_beach"
3996
+ ]
3997
+ },
3998
+ {
3999
+ "file": "syn/train/data-00351-of-00451.parquet",
4000
+ "rows": 1000,
4001
+ "bytes": 1882213379,
4002
+ "classes": {
4003
+ "handrail": 1000
4004
+ },
4005
+ "environments": [
4006
+ "miami_beach",
4007
+ "miami_beach_night"
4008
+ ]
4009
+ },
4010
+ {
4011
+ "file": "syn/train/data-00352-of-00451.parquet",
4012
+ "rows": 1000,
4013
+ "bytes": 2015179819,
4014
+ "classes": {
4015
+ "handrail": 1000
4016
+ },
4017
+ "environments": [
4018
+ "miami_beach_night"
4019
+ ]
4020
+ },
4021
+ {
4022
+ "file": "syn/train/data-00353-of-00451.parquet",
4023
+ "rows": 1000,
4024
+ "bytes": 1660747641,
4025
+ "classes": {
4026
+ "handrail": 1000
4027
+ },
4028
+ "environments": [
4029
+ "miami_beach_night",
4030
+ "night_city"
4031
+ ]
4032
+ },
4033
+ {
4034
+ "file": "syn/train/data-00354-of-00451.parquet",
4035
+ "rows": 1000,
4036
+ "bytes": 1720811607,
4037
+ "classes": {
4038
+ "handrail": 1000
4039
+ },
4040
+ "environments": [
4041
+ "night_city"
4042
+ ]
4043
+ },
4044
+ {
4045
+ "file": "syn/train/data-00355-of-00451.parquet",
4046
+ "rows": 1000,
4047
+ "bytes": 1661927379,
4048
+ "classes": {
4049
+ "handrail": 1000
4050
+ },
4051
+ "environments": [
4052
+ "night_city"
4053
+ ]
4054
+ },
4055
+ {
4056
+ "file": "syn/train/data-00356-of-00451.parquet",
4057
+ "rows": 1000,
4058
+ "bytes": 2186706194,
4059
+ "classes": {
4060
+ "handrail": 1000
4061
+ },
4062
+ "environments": [
4063
+ "night_city",
4064
+ "nordic_harbour"
4065
+ ]
4066
+ },
4067
+ {
4068
+ "file": "syn/train/data-00357-of-00451.parquet",
4069
+ "rows": 1000,
4070
+ "bytes": 1981468795,
4071
+ "classes": {
4072
+ "handrail": 1000
4073
+ },
4074
+ "environments": [
4075
+ "nordic_harbour"
4076
+ ]
4077
+ },
4078
+ {
4079
+ "file": "syn/train/data-00358-of-00451.parquet",
4080
+ "rows": 1000,
4081
+ "bytes": 1870737139,
4082
+ "classes": {
4083
+ "handrail": 1000
4084
+ },
4085
+ "environments": [
4086
+ "nordic_harbour",
4087
+ "nordic_harbour_night"
4088
+ ]
4089
+ },
4090
+ {
4091
+ "file": "syn/train/data-00359-of-00451.parquet",
4092
+ "rows": 1000,
4093
+ "bytes": 1521296576,
4094
+ "classes": {
4095
+ "handrail": 1000
4096
+ },
4097
+ "environments": [
4098
+ "nordic_harbour_night"
4099
+ ]
4100
+ },
4101
+ {
4102
+ "file": "syn/train/data-00360-of-00451.parquet",
4103
+ "rows": 1000,
4104
+ "bytes": 2125559371,
4105
+ "classes": {
4106
+ "handrail": 1000
4107
+ },
4108
+ "environments": [
4109
+ "nordic_harbour_night",
4110
+ "nordic_harbour_sunset"
4111
+ ]
4112
+ },
4113
+ {
4114
+ "file": "syn/train/data-00361-of-00451.parquet",
4115
+ "rows": 1000,
4116
+ "bytes": 1903901501,
4117
+ "classes": {
4118
+ "handrail": 1000
4119
+ },
4120
+ "environments": [
4121
+ "nordic_harbour_sunset"
4122
+ ]
4123
+ },
4124
+ {
4125
+ "file": "syn/train/data-00362-of-00451.parquet",
4126
+ "rows": 1000,
4127
+ "bytes": 2194145168,
4128
+ "classes": {
4129
+ "handrail": 1000
4130
+ },
4131
+ "environments": [
4132
+ "nordic_harbour_sunset",
4133
+ "stylized_city"
4134
+ ]
4135
+ },
4136
+ {
4137
+ "file": "syn/train/data-00363-of-00451.parquet",
4138
+ "rows": 1000,
4139
+ "bytes": 1933219944,
4140
+ "classes": {
4141
+ "handrail": 1000
4142
+ },
4143
+ "environments": [
4144
+ "stylized_city"
4145
+ ]
4146
+ },
4147
+ {
4148
+ "file": "syn/train/data-00364-of-00451.parquet",
4149
+ "rows": 1000,
4150
+ "bytes": 1935496255,
4151
+ "classes": {
4152
+ "handrail": 1000
4153
+ },
4154
+ "environments": [
4155
+ "stylized_city",
4156
+ "stylized_city_night"
4157
+ ]
4158
+ },
4159
+ {
4160
+ "file": "syn/train/data-00365-of-00451.parquet",
4161
+ "rows": 1000,
4162
+ "bytes": 1818428752,
4163
+ "classes": {
4164
+ "handrail": 1000
4165
+ },
4166
+ "environments": [
4167
+ "stylized_city_night"
4168
+ ]
4169
+ },
4170
+ {
4171
+ "file": "syn/train/data-00366-of-00451.parquet",
4172
+ "rows": 1000,
4173
+ "bytes": 1749871134,
4174
+ "classes": {
4175
+ "handrail": 1000
4176
+ },
4177
+ "environments": [
4178
+ "stylized_city_night",
4179
+ "stylized_city_sunset"
4180
+ ]
4181
+ },
4182
+ {
4183
+ "file": "syn/train/data-00367-of-00451.parquet",
4184
+ "rows": 1000,
4185
+ "bytes": 2107212275,
4186
+ "classes": {
4187
+ "handrail": 1000
4188
+ },
4189
+ "environments": [
4190
+ "stylized_city_sunset"
4191
+ ]
4192
+ },
4193
+ {
4194
+ "file": "syn/train/data-00368-of-00451.parquet",
4195
+ "rows": 1000,
4196
+ "bytes": 1850214227,
4197
+ "classes": {
4198
+ "handrail": 1000
4199
+ },
4200
+ "environments": [
4201
+ "stylized_city_sunset",
4202
+ "suburbs"
4203
+ ]
4204
+ },
4205
+ {
4206
+ "file": "syn/train/data-00369-of-00451.parquet",
4207
+ "rows": 1000,
4208
+ "bytes": 1908578264,
4209
+ "classes": {
4210
+ "handrail": 1000
4211
+ },
4212
+ "environments": [
4213
+ "suburbs"
4214
+ ]
4215
+ },
4216
+ {
4217
+ "file": "syn/train/data-00370-of-00451.parquet",
4218
+ "rows": 1000,
4219
+ "bytes": 1813401209,
4220
+ "classes": {
4221
+ "handrail": 1000
4222
+ },
4223
+ "environments": [
4224
+ "suburbs",
4225
+ "suburbs_night"
4226
+ ]
4227
+ },
4228
+ {
4229
+ "file": "syn/train/data-00371-of-00451.parquet",
4230
+ "rows": 1000,
4231
+ "bytes": 1737450072,
4232
+ "classes": {
4233
+ "handrail": 1000
4234
+ },
4235
+ "environments": [
4236
+ "suburbs_night",
4237
+ "suburbs_sunset"
4238
+ ]
4239
+ },
4240
+ {
4241
+ "file": "syn/train/data-00372-of-00451.parquet",
4242
+ "rows": 1000,
4243
+ "bytes": 1838622582,
4244
+ "classes": {
4245
+ "handrail": 1000
4246
+ },
4247
+ "environments": [
4248
+ "suburbs_sunset"
4249
+ ]
4250
+ },
4251
+ {
4252
+ "file": "syn/train/data-00373-of-00451.parquet",
4253
+ "rows": 1000,
4254
+ "bytes": 1950635742,
4255
+ "classes": {
4256
+ "handrail": 1000
4257
+ },
4258
+ "environments": [
4259
+ "suburbs_sunset",
4260
+ "tokyo_street"
4261
+ ]
4262
+ },
4263
+ {
4264
+ "file": "syn/train/data-00374-of-00451.parquet",
4265
+ "rows": 1000,
4266
+ "bytes": 1923506797,
4267
+ "classes": {
4268
+ "handrail": 1000
4269
+ },
4270
+ "environments": [
4271
+ "tokyo_street",
4272
+ "tokyo_street_sunset"
4273
+ ]
4274
+ },
4275
+ {
4276
+ "file": "syn/train/data-00375-of-00451.parquet",
4277
+ "rows": 1000,
4278
+ "bytes": 1915076827,
4279
+ "classes": {
4280
+ "handrail": 1000
4281
+ },
4282
+ "environments": [
4283
+ "tokyo_street_sunset"
4284
+ ]
4285
+ },
4286
+ {
4287
+ "file": "syn/train/data-00376-of-00451.parquet",
4288
+ "rows": 1000,
4289
+ "bytes": 1862443661,
4290
+ "classes": {
4291
+ "handrail": 1000
4292
+ },
4293
+ "environments": [
4294
+ "tokyo_street_sunset",
4295
+ "village"
4296
+ ]
4297
+ },
4298
+ {
4299
+ "file": "syn/train/data-00377-of-00451.parquet",
4300
+ "rows": 1000,
4301
+ "bytes": 1827956703,
4302
+ "classes": {
4303
+ "handrail": 1000
4304
+ },
4305
+ "environments": [
4306
+ "village"
4307
+ ]
4308
+ },
4309
+ {
4310
+ "file": "syn/train/data-00378-of-00451.parquet",
4311
+ "rows": 1000,
4312
+ "bytes": 1822560591,
4313
+ "classes": {
4314
+ "handrail": 1000
4315
+ },
4316
+ "environments": [
4317
+ "village",
4318
+ "village_sunset"
4319
+ ]
4320
+ },
4321
+ {
4322
+ "file": "syn/train/data-00379-of-00451.parquet",
4323
+ "rows": 1000,
4324
+ "bytes": 1940601672,
4325
+ "classes": {
4326
+ "handrail": 1000
4327
+ },
4328
+ "environments": [
4329
+ "village_sunset"
4330
+ ]
4331
+ },
4332
+ {
4333
+ "file": "syn/train/data-00380-of-00451.parquet",
4334
+ "rows": 1000,
4335
+ "bytes": 1822498632,
4336
+ "classes": {
4337
+ "handrail": 1000
4338
+ },
4339
+ "environments": [
4340
+ "village_sunset"
4341
+ ]
4342
+ },
4343
+ {
4344
+ "file": "syn/train/data-00381-of-00451.parquet",
4345
+ "rows": 1000,
4346
+ "bytes": 2029377659,
4347
+ "classes": {
4348
+ "handrail": 200,
4349
+ "pedestrian_signal": 800
4350
+ },
4351
+ "environments": [
4352
+ "downtown_west",
4353
+ "village_sunset"
4354
+ ]
4355
+ },
4356
+ {
4357
+ "file": "syn/train/data-00382-of-00451.parquet",
4358
+ "rows": 1000,
4359
+ "bytes": 2554646531,
4360
+ "classes": {
4361
+ "pedestrian_signal": 1000
4362
+ },
4363
+ "environments": [
4364
+ "downtown_west"
4365
+ ]
4366
+ },
4367
+ {
4368
+ "file": "syn/train/data-00383-of-00451.parquet",
4369
+ "rows": 1000,
4370
+ "bytes": 2300396190,
4371
+ "classes": {
4372
+ "pedestrian_signal": 1000
4373
+ },
4374
+ "environments": [
4375
+ "downtown_west"
4376
+ ]
4377
+ },
4378
+ {
4379
+ "file": "syn/train/data-00384-of-00451.parquet",
4380
+ "rows": 1000,
4381
+ "bytes": 2145199006,
4382
+ "classes": {
4383
+ "pedestrian_signal": 1000
4384
+ },
4385
+ "environments": [
4386
+ "downtown_west",
4387
+ "downtown_west_night"
4388
+ ]
4389
+ },
4390
+ {
4391
+ "file": "syn/train/data-00385-of-00451.parquet",
4392
+ "rows": 1000,
4393
+ "bytes": 1996941568,
4394
+ "classes": {
4395
+ "pedestrian_signal": 1000
4396
+ },
4397
+ "environments": [
4398
+ "downtown_west_night"
4399
+ ]
4400
+ },
4401
+ {
4402
+ "file": "syn/train/data-00386-of-00451.parquet",
4403
+ "rows": 1000,
4404
+ "bytes": 1820181186,
4405
+ "classes": {
4406
+ "pedestrian_signal": 1000
4407
+ },
4408
+ "environments": [
4409
+ "downtown_west_night"
4410
+ ]
4411
+ },
4412
+ {
4413
+ "file": "syn/train/data-00387-of-00451.parquet",
4414
+ "rows": 1000,
4415
+ "bytes": 1883259722,
4416
+ "classes": {
4417
+ "pedestrian_signal": 1000
4418
+ },
4419
+ "environments": [
4420
+ "downtown_west_night",
4421
+ "downtown_west_sunset"
4422
+ ]
4423
+ },
4424
+ {
4425
+ "file": "syn/train/data-00388-of-00451.parquet",
4426
+ "rows": 1000,
4427
+ "bytes": 2184029278,
4428
+ "classes": {
4429
+ "pedestrian_signal": 1000
4430
+ },
4431
+ "environments": [
4432
+ "downtown_west_sunset"
4433
+ ]
4434
+ },
4435
+ {
4436
+ "file": "syn/train/data-00389-of-00451.parquet",
4437
+ "rows": 1000,
4438
+ "bytes": 2415127740,
4439
+ "classes": {
4440
+ "pedestrian_signal": 1000
4441
+ },
4442
+ "environments": [
4443
+ "downtown_west_sunset"
4444
+ ]
4445
+ },
4446
+ {
4447
+ "file": "syn/train/data-00390-of-00451.parquet",
4448
+ "rows": 1000,
4449
+ "bytes": 2414148943,
4450
+ "classes": {
4451
+ "pedestrian_signal": 1000
4452
+ },
4453
+ "environments": [
4454
+ "downtown_west_sunset"
4455
+ ]
4456
+ },
4457
+ {
4458
+ "file": "syn/train/data-00391-of-00451.parquet",
4459
+ "rows": 1000,
4460
+ "bytes": 2021797286,
4461
+ "classes": {
4462
+ "pedestrian_signal": 1000
4463
+ },
4464
+ "environments": [
4465
+ "downtown_west_sunset",
4466
+ "hong_kong_street"
4467
+ ]
4468
+ },
4469
+ {
4470
+ "file": "syn/train/data-00392-of-00451.parquet",
4471
+ "rows": 1000,
4472
+ "bytes": 1881027761,
4473
+ "classes": {
4474
+ "pedestrian_signal": 1000
4475
+ },
4476
+ "environments": [
4477
+ "hong_kong_street",
4478
+ "hong_kong_street_sunset"
4479
+ ]
4480
+ },
4481
+ {
4482
+ "file": "syn/train/data-00393-of-00451.parquet",
4483
+ "rows": 1000,
4484
+ "bytes": 1894786862,
4485
+ "classes": {
4486
+ "pedestrian_signal": 1000
4487
+ },
4488
+ "environments": [
4489
+ "hong_kong_street_sunset",
4490
+ "miami_beach"
4491
+ ]
4492
+ },
4493
+ {
4494
+ "file": "syn/train/data-00394-of-00451.parquet",
4495
+ "rows": 1000,
4496
+ "bytes": 2992263296,
4497
+ "classes": {
4498
+ "pedestrian_signal": 1000
4499
+ },
4500
+ "environments": [
4501
+ "miami_beach"
4502
+ ]
4503
+ },
4504
+ {
4505
+ "file": "syn/train/data-00395-of-00451.parquet",
4506
+ "rows": 1000,
4507
+ "bytes": 2919152478,
4508
+ "classes": {
4509
+ "pedestrian_signal": 1000
4510
+ },
4511
+ "environments": [
4512
+ "miami_beach",
4513
+ "miami_beach_night"
4514
+ ]
4515
+ },
4516
+ {
4517
+ "file": "syn/train/data-00396-of-00451.parquet",
4518
+ "rows": 1000,
4519
+ "bytes": 2208326641,
4520
+ "classes": {
4521
+ "pedestrian_signal": 1000
4522
+ },
4523
+ "environments": [
4524
+ "miami_beach_night",
4525
+ "night_city"
4526
+ ]
4527
+ },
4528
+ {
4529
+ "file": "syn/train/data-00397-of-00451.parquet",
4530
+ "rows": 1000,
4531
+ "bytes": 2010418843,
4532
+ "classes": {
4533
+ "pedestrian_signal": 1000
4534
+ },
4535
+ "environments": [
4536
+ "night_city",
4537
+ "nordic_harbour"
4538
+ ]
4539
+ },
4540
+ {
4541
+ "file": "syn/train/data-00398-of-00451.parquet",
4542
+ "rows": 1000,
4543
+ "bytes": 2140453227,
4544
+ "classes": {
4545
+ "pedestrian_signal": 1000
4546
+ },
4547
+ "environments": [
4548
+ "nordic_harbour"
4549
+ ]
4550
+ },
4551
+ {
4552
+ "file": "syn/train/data-00399-of-00451.parquet",
4553
+ "rows": 1000,
4554
+ "bytes": 1685225551,
4555
+ "classes": {
4556
+ "pedestrian_signal": 1000
4557
+ },
4558
+ "environments": [
4559
+ "nordic_harbour",
4560
+ "nordic_harbour_night"
4561
+ ]
4562
+ },
4563
+ {
4564
+ "file": "syn/train/data-00400-of-00451.parquet",
4565
+ "rows": 1000,
4566
+ "bytes": 1673892612,
4567
+ "classes": {
4568
+ "pedestrian_signal": 1000
4569
+ },
4570
+ "environments": [
4571
+ "nordic_harbour_night"
4572
+ ]
4573
+ },
4574
+ {
4575
+ "file": "syn/train/data-00401-of-00451.parquet",
4576
+ "rows": 1000,
4577
+ "bytes": 1620117711,
4578
+ "classes": {
4579
+ "pedestrian_signal": 1000
4580
+ },
4581
+ "environments": [
4582
+ "nordic_harbour_night",
4583
+ "nordic_harbour_sunset"
4584
+ ]
4585
+ },
4586
+ {
4587
+ "file": "syn/train/data-00402-of-00451.parquet",
4588
+ "rows": 1000,
4589
+ "bytes": 2093228759,
4590
+ "classes": {
4591
+ "pedestrian_signal": 1000
4592
+ },
4593
+ "environments": [
4594
+ "nordic_harbour_sunset"
4595
+ ]
4596
+ },
4597
+ {
4598
+ "file": "syn/train/data-00403-of-00451.parquet",
4599
+ "rows": 1000,
4600
+ "bytes": 1580288258,
4601
+ "classes": {
4602
+ "pedestrian_signal": 1000
4603
+ },
4604
+ "environments": [
4605
+ "nordic_harbour_sunset",
4606
+ "stylized_city"
4607
+ ]
4608
+ },
4609
+ {
4610
+ "file": "syn/train/data-00404-of-00451.parquet",
4611
+ "rows": 1000,
4612
+ "bytes": 1573823203,
4613
+ "classes": {
4614
+ "pedestrian_signal": 1000
4615
+ },
4616
+ "environments": [
4617
+ "stylized_city"
4618
+ ]
4619
+ },
4620
+ {
4621
+ "file": "syn/train/data-00405-of-00451.parquet",
4622
+ "rows": 1000,
4623
+ "bytes": 1456675512,
4624
+ "classes": {
4625
+ "pedestrian_signal": 1000
4626
+ },
4627
+ "environments": [
4628
+ "stylized_city",
4629
+ "stylized_city_night"
4630
+ ]
4631
+ },
4632
+ {
4633
+ "file": "syn/train/data-00406-of-00451.parquet",
4634
+ "rows": 1000,
4635
+ "bytes": 1338314107,
4636
+ "classes": {
4637
+ "pedestrian_signal": 1000
4638
+ },
4639
+ "environments": [
4640
+ "stylized_city_night"
4641
+ ]
4642
+ },
4643
+ {
4644
+ "file": "syn/train/data-00407-of-00451.parquet",
4645
+ "rows": 1000,
4646
+ "bytes": 1446813401,
4647
+ "classes": {
4648
+ "pedestrian_signal": 1000
4649
+ },
4650
+ "environments": [
4651
+ "stylized_city_night",
4652
+ "stylized_city_sunset"
4653
+ ]
4654
+ },
4655
+ {
4656
+ "file": "syn/train/data-00408-of-00451.parquet",
4657
+ "rows": 1000,
4658
+ "bytes": 1535244123,
4659
+ "classes": {
4660
+ "pedestrian_signal": 1000
4661
+ },
4662
+ "environments": [
4663
+ "stylized_city_sunset"
4664
+ ]
4665
+ },
4666
+ {
4667
+ "file": "syn/train/data-00409-of-00451.parquet",
4668
+ "rows": 1000,
4669
+ "bytes": 2168841510,
4670
+ "classes": {
4671
+ "pedestrian_signal": 1000
4672
+ },
4673
+ "environments": [
4674
+ "stylized_city_sunset",
4675
+ "suburbs"
4676
+ ]
4677
+ },
4678
+ {
4679
+ "file": "syn/train/data-00410-of-00451.parquet",
4680
+ "rows": 1000,
4681
+ "bytes": 2590160093,
4682
+ "classes": {
4683
+ "pedestrian_signal": 1000
4684
+ },
4685
+ "environments": [
4686
+ "suburbs"
4687
+ ]
4688
+ },
4689
+ {
4690
+ "file": "syn/train/data-00411-of-00451.parquet",
4691
+ "rows": 1000,
4692
+ "bytes": 2313932748,
4693
+ "classes": {
4694
+ "pedestrian_signal": 1000
4695
+ },
4696
+ "environments": [
4697
+ "suburbs",
4698
+ "suburbs_night"
4699
+ ]
4700
+ },
4701
+ {
4702
+ "file": "syn/train/data-00412-of-00451.parquet",
4703
+ "rows": 1000,
4704
+ "bytes": 2331047212,
4705
+ "classes": {
4706
+ "pedestrian_signal": 1000
4707
+ },
4708
+ "environments": [
4709
+ "suburbs_night"
4710
+ ]
4711
+ },
4712
+ {
4713
+ "file": "syn/train/data-00413-of-00451.parquet",
4714
+ "rows": 1000,
4715
+ "bytes": 2217684374,
4716
+ "classes": {
4717
+ "pedestrian_signal": 1000
4718
+ },
4719
+ "environments": [
4720
+ "suburbs_night",
4721
+ "suburbs_sunset"
4722
+ ]
4723
+ },
4724
+ {
4725
+ "file": "syn/train/data-00414-of-00451.parquet",
4726
+ "rows": 1000,
4727
+ "bytes": 2515977537,
4728
+ "classes": {
4729
+ "pedestrian_signal": 1000
4730
+ },
4731
+ "environments": [
4732
+ "suburbs_sunset"
4733
+ ]
4734
+ },
4735
+ {
4736
+ "file": "syn/train/data-00415-of-00451.parquet",
4737
+ "rows": 1000,
4738
+ "bytes": 1946540285,
4739
+ "classes": {
4740
+ "pedestrian_signal": 1000
4741
+ },
4742
+ "environments": [
4743
+ "suburbs_sunset",
4744
+ "tokyo_street"
4745
+ ]
4746
+ },
4747
+ {
4748
+ "file": "syn/train/data-00416-of-00451.parquet",
4749
+ "rows": 1000,
4750
+ "bytes": 1851060487,
4751
+ "classes": {
4752
+ "pedestrian_signal": 1000
4753
+ },
4754
+ "environments": [
4755
+ "tokyo_street"
4756
+ ]
4757
+ },
4758
+ {
4759
+ "file": "syn/train/data-00417-of-00451.parquet",
4760
+ "rows": 1000,
4761
+ "bytes": 1942407570,
4762
+ "classes": {
4763
+ "pedestrian_signal": 1000
4764
+ },
4765
+ "environments": [
4766
+ "tokyo_street",
4767
+ "tokyo_street_sunset"
4768
+ ]
4769
+ },
4770
+ {
4771
+ "file": "syn/train/data-00418-of-00451.parquet",
4772
+ "rows": 1000,
4773
+ "bytes": 1899289594,
4774
+ "classes": {
4775
+ "pedestrian_signal": 1000
4776
+ },
4777
+ "environments": [
4778
+ "tokyo_street_sunset"
4779
+ ]
4780
+ },
4781
+ {
4782
+ "file": "syn/train/data-00419-of-00451.parquet",
4783
+ "rows": 1000,
4784
+ "bytes": 2157251955,
4785
+ "classes": {
4786
+ "pedestrian_signal": 1000
4787
+ },
4788
+ "environments": [
4789
+ "tokyo_street_sunset",
4790
+ "village"
4791
+ ]
4792
+ },
4793
+ {
4794
+ "file": "syn/train/data-00420-of-00451.parquet",
4795
+ "rows": 1000,
4796
+ "bytes": 2219207807,
4797
+ "classes": {
4798
+ "pedestrian_signal": 1000
4799
+ },
4800
+ "environments": [
4801
+ "village"
4802
+ ]
4803
+ },
4804
+ {
4805
+ "file": "syn/train/data-00421-of-00451.parquet",
4806
+ "rows": 1000,
4807
+ "bytes": 2777330957,
4808
+ "classes": {
4809
+ "pedestrian_signal": 1000
4810
+ },
4811
+ "environments": [
4812
+ "village"
4813
+ ]
4814
+ },
4815
+ {
4816
+ "file": "syn/train/data-00422-of-00451.parquet",
4817
+ "rows": 1000,
4818
+ "bytes": 2611946132,
4819
+ "classes": {
4820
+ "pedestrian_signal": 1000
4821
+ },
4822
+ "environments": [
4823
+ "village",
4824
+ "village_sunset"
4825
+ ]
4826
+ },
4827
+ {
4828
+ "file": "syn/train/data-00423-of-00451.parquet",
4829
+ "rows": 1000,
4830
+ "bytes": 2252407453,
4831
+ "classes": {
4832
+ "pedestrian_signal": 1000
4833
+ },
4834
+ "environments": [
4835
+ "village_sunset"
4836
+ ]
4837
+ },
4838
+ {
4839
+ "file": "syn/train/data-00424-of-00451.parquet",
4840
+ "rows": 1000,
4841
+ "bytes": 2687244363,
4842
+ "classes": {
4843
+ "pedestrian_signal": 1000
4844
+ },
4845
+ "environments": [
4846
+ "village_sunset"
4847
+ ]
4848
+ },
4849
+ {
4850
+ "file": "syn/train/data-00425-of-00451.parquet",
4851
+ "rows": 1000,
4852
+ "bytes": 2563033540,
4853
+ "classes": {
4854
+ "pedestrian_signal": 850,
4855
+ "revolving_door": 150
4856
+ },
4857
+ "environments": [
4858
+ "airport",
4859
+ "village_sunset"
4860
+ ]
4861
+ },
4862
+ {
4863
+ "file": "syn/train/data-00426-of-00451.parquet",
4864
+ "rows": 1000,
4865
+ "bytes": 1239287379,
4866
+ "classes": {
4867
+ "revolving_door": 1000
4868
+ },
4869
+ "environments": [
4870
+ "airport"
4871
+ ]
4872
+ },
4873
+ {
4874
+ "file": "syn/train/data-00427-of-00451.parquet",
4875
+ "rows": 1000,
4876
+ "bytes": 1869793029,
4877
+ "classes": {
4878
+ "revolving_door": 1000
4879
+ },
4880
+ "environments": [
4881
+ "airport",
4882
+ "downtown_west"
4883
+ ]
4884
+ },
4885
+ {
4886
+ "file": "syn/train/data-00428-of-00451.parquet",
4887
+ "rows": 1000,
4888
+ "bytes": 1537835104,
4889
+ "classes": {
4890
+ "revolving_door": 1000
4891
+ },
4892
+ "environments": [
4893
+ "downtown_west"
4894
+ ]
4895
+ },
4896
+ {
4897
+ "file": "syn/train/data-00429-of-00451.parquet",
4898
+ "rows": 1000,
4899
+ "bytes": 1537896874,
4900
+ "classes": {
4901
+ "revolving_door": 1000
4902
+ },
4903
+ "environments": [
4904
+ "downtown_west"
4905
+ ]
4906
+ },
4907
+ {
4908
+ "file": "syn/train/data-00430-of-00451.parquet",
4909
+ "rows": 1000,
4910
+ "bytes": 1639375696,
4911
+ "classes": {
4912
+ "revolving_door": 1000
4913
+ },
4914
+ "environments": [
4915
+ "downtown_west",
4916
+ "downtown_west_night"
4917
+ ]
4918
+ },
4919
+ {
4920
+ "file": "syn/train/data-00431-of-00451.parquet",
4921
+ "rows": 1000,
4922
+ "bytes": 1545739487,
4923
+ "classes": {
4924
+ "revolving_door": 1000
4925
+ },
4926
+ "environments": [
4927
+ "downtown_west_night"
4928
+ ]
4929
+ },
4930
+ {
4931
+ "file": "syn/train/data-00432-of-00451.parquet",
4932
+ "rows": 1000,
4933
+ "bytes": 1259026069,
4934
+ "classes": {
4935
+ "revolving_door": 1000
4936
+ },
4937
+ "environments": [
4938
+ "downtown_west_night"
4939
+ ]
4940
+ },
4941
+ {
4942
+ "file": "syn/train/data-00433-of-00451.parquet",
4943
+ "rows": 1000,
4944
+ "bytes": 1411966250,
4945
+ "classes": {
4946
+ "revolving_door": 1000
4947
+ },
4948
+ "environments": [
4949
+ "downtown_west_night"
4950
+ ]
4951
+ },
4952
+ {
4953
+ "file": "syn/train/data-00434-of-00451.parquet",
4954
+ "rows": 1000,
4955
+ "bytes": 1780623789,
4956
+ "classes": {
4957
+ "revolving_door": 1000
4958
+ },
4959
+ "environments": [
4960
+ "downtown_west_night",
4961
+ "downtown_west_sunset"
4962
+ ]
4963
+ },
4964
+ {
4965
+ "file": "syn/train/data-00435-of-00451.parquet",
4966
+ "rows": 1000,
4967
+ "bytes": 1440000705,
4968
+ "classes": {
4969
+ "revolving_door": 1000
4970
+ },
4971
+ "environments": [
4972
+ "downtown_west_sunset"
4973
+ ]
4974
+ },
4975
+ {
4976
+ "file": "syn/train/data-00436-of-00451.parquet",
4977
+ "rows": 1000,
4978
+ "bytes": 1440313392,
4979
+ "classes": {
4980
+ "revolving_door": 1000
4981
+ },
4982
+ "environments": [
4983
+ "downtown_west_sunset"
4984
+ ]
4985
+ },
4986
+ {
4987
+ "file": "syn/train/data-00437-of-00451.parquet",
4988
+ "rows": 1000,
4989
+ "bytes": 1552065414,
4990
+ "classes": {
4991
+ "revolving_door": 1000
4992
+ },
4993
+ "environments": [
4994
+ "downtown_west_sunset",
4995
+ "lobby"
4996
+ ]
4997
+ },
4998
+ {
4999
+ "file": "syn/train/data-00438-of-00451.parquet",
5000
+ "rows": 1000,
5001
+ "bytes": 1390511092,
5002
+ "classes": {
5003
+ "revolving_door": 1000
5004
+ },
5005
+ "environments": [
5006
+ "lobby"
5007
+ ]
5008
+ },
5009
+ {
5010
+ "file": "syn/train/data-00439-of-00451.parquet",
5011
+ "rows": 1000,
5012
+ "bytes": 1563041921,
5013
+ "classes": {
5014
+ "revolving_door": 1000
5015
+ },
5016
+ "environments": [
5017
+ "lobby",
5018
+ "miami_beach"
5019
+ ]
5020
+ },
5021
+ {
5022
+ "file": "syn/train/data-00440-of-00451.parquet",
5023
+ "rows": 1000,
5024
+ "bytes": 1556166494,
5025
+ "classes": {
5026
+ "revolving_door": 1000
5027
+ },
5028
+ "environments": [
5029
+ "miami_beach"
5030
+ ]
5031
+ },
5032
+ {
5033
+ "file": "syn/train/data-00441-of-00451.parquet",
5034
+ "rows": 1000,
5035
+ "bytes": 1571499022,
5036
+ "classes": {
5037
+ "revolving_door": 1000
5038
+ },
5039
+ "environments": [
5040
+ "miami_beach",
5041
+ "miami_beach_night"
5042
+ ]
5043
+ },
5044
+ {
5045
+ "file": "syn/train/data-00442-of-00451.parquet",
5046
+ "rows": 1000,
5047
+ "bytes": 1422263488,
5048
+ "classes": {
5049
+ "revolving_door": 1000
5050
+ },
5051
+ "environments": [
5052
+ "miami_beach_night"
5053
+ ]
5054
+ },
5055
+ {
5056
+ "file": "syn/train/data-00443-of-00451.parquet",
5057
+ "rows": 1000,
5058
+ "bytes": 1614539813,
5059
+ "classes": {
5060
+ "revolving_door": 1000
5061
+ },
5062
+ "environments": [
5063
+ "miami_beach_night",
5064
+ "shopping_mall"
5065
+ ]
5066
+ },
5067
+ {
5068
+ "file": "syn/train/data-00444-of-00451.parquet",
5069
+ "rows": 1000,
5070
+ "bytes": 1571799069,
5071
+ "classes": {
5072
+ "revolving_door": 1000
5073
+ },
5074
+ "environments": [
5075
+ "shopping_mall"
5076
+ ]
5077
+ },
5078
+ {
5079
+ "file": "syn/train/data-00445-of-00451.parquet",
5080
+ "rows": 1000,
5081
+ "bytes": 1602807192,
5082
+ "classes": {
5083
+ "revolving_door": 1000
5084
+ },
5085
+ "environments": [
5086
+ "shopping_mall"
5087
+ ]
5088
+ },
5089
+ {
5090
+ "file": "syn/train/data-00446-of-00451.parquet",
5091
+ "rows": 1000,
5092
+ "bytes": 1394399229,
5093
+ "classes": {
5094
+ "revolving_door": 1000
5095
+ },
5096
+ "environments": [
5097
+ "shopping_mall",
5098
+ "skyscraper"
5099
+ ]
5100
+ },
5101
+ {
5102
+ "file": "syn/train/data-00447-of-00451.parquet",
5103
+ "rows": 1000,
5104
+ "bytes": 1242066181,
5105
+ "classes": {
5106
+ "revolving_door": 1000
5107
+ },
5108
+ "environments": [
5109
+ "skyscraper",
5110
+ "subway_station"
5111
+ ]
5112
+ },
5113
+ {
5114
+ "file": "syn/train/data-00448-of-00451.parquet",
5115
+ "rows": 1000,
5116
+ "bytes": 1340575335,
5117
+ "classes": {
5118
+ "revolving_door": 1000
5119
+ },
5120
+ "environments": [
5121
+ "subway_station"
5122
+ ]
5123
+ },
5124
+ {
5125
+ "file": "syn/train/data-00449-of-00451.parquet",
5126
+ "rows": 1000,
5127
+ "bytes": 1385294156,
5128
+ "classes": {
5129
+ "revolving_door": 1000
5130
+ },
5131
+ "environments": [
5132
+ "subway_station",
5133
+ "train_terminal"
5134
+ ]
5135
+ },
5136
+ {
5137
+ "file": "syn/train/data-00450-of-00451.parquet",
5138
+ "rows": 860,
5139
+ "bytes": 1178999634,
5140
+ "classes": {
5141
+ "revolving_door": 860
5142
+ },
5143
+ "environments": [
5144
+ "train_terminal"
5145
+ ]
5146
+ }
5147
+ ]
5148
+ }
tools/README.md ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # NavAble tools
2
+
3
+ Small, dependency-light helpers for the [NavAble dataset](https://huggingface.co/datasets/NavAble/NeurIPS_2026_BLV).
4
+
5
+ ```bash
6
+ hf download NavAble/NeurIPS_2026_BLV --repo-type dataset --include "tools/*" --local-dir navable
7
+ pip install -r navable/tools/requirements.txt
8
+ ```
9
+
10
+ | Script | What it does |
11
+ |---|---|
12
+ | `navable.py` | Importable helpers: class names and palette, loaders (`load`, `load_syn` for class subsets of `syn`), depth / mask / box decoding and visualization, the paper's metrics, label remapping for the released checkpoints, a PyTorch `Dataset` wrapper. |
13
+ | `download.py` | Download only the parts you need (real data, `syn` shards of chosen classes, 3D assets, checkpoints); prints sizes first. |
14
+ | `explore.py` | Save a grid of random samples: RGB, class mask and, for `syn`, depth, surface normals and 2D boxes. |
15
+ | `export.py` | Write a config/split as image + label folders (mmsegmentation layout), optionally with depth and normals. |
16
+ | `evaluate.py` | Score prediction PNGs on `real_ours` with the paper's mIoU / precision / recall. |
17
+ | `eval_checkpoint.py` | Run a released mmsegmentation checkpoint (DeepLabv3+, EncNet, SegFormer, Mask2Former) on `real_ours/test`. |
18
+
19
+ Every script has `--help` with examples. All of them read only the parquet shards they need,
20
+ straight from the Hub, or from a downloaded copy with `--data-dir`.
tools/download.py ADDED
@@ -0,0 +1,105 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ """Download just the parts of NavAble you need, keeping the repo layout.
3
+
4
+ The full repo is about 985 GB (syn alone is 816 GB), so pick what you need. The plan with file
5
+ counts and sizes is printed first; nothing is downloaded with --dry-run.
6
+
7
+ Examples:
8
+ python tools/download.py --real # real_ours, all splits (1.5 GB)
9
+ python tools/download.py --real --curated # + real_curated (114 GB)
10
+ python tools/download.py --syn escalator handrail # only the syn shards holding these classes
11
+ python tools/download.py --syn all --max-shards 5 # a small syn sample (about 10 GB)
12
+ python tools/download.py --assets door_button # 3D assets (.glb/.ply/.usdz) of one class
13
+ python tools/download.py --checkpoints segformer --train-data real+syn
14
+ python tools/download.py --checkpoints all --dry-run # list every checkpoint and its size
15
+
16
+ The download folder (--local-dir, default ./navable) can then be read with
17
+ `load_dataset("./navable", "real_ours")`, or by the other tools with --data-dir ./navable.
18
+ """
19
+ from __future__ import annotations
20
+
21
+ import argparse
22
+ import re
23
+ import sys
24
+ from collections import defaultdict
25
+ from pathlib import Path
26
+
27
+ sys.path.insert(0, str(Path(__file__).resolve().parent))
28
+ import navable as nv # noqa: E402
29
+
30
+ TRAIN_DATA = ["real", "real+cur", "real+syn", "real+cur+syn"]
31
+ CKPT_RE = re.compile(r"^navable_(?P<model>[a-z0-9]+(?:-headonly)?)(?:-vit-b16)?[-_](?P<data>real(?:[-+]cur)?(?:[-+]syn)?)\.pth$")
32
+ ALWAYS = ["README.md", "class_index.json", "syn/shard_index.json", "docs/datasheet.md"]
33
+
34
+
35
+ def checkpoint_key(path: str) -> tuple[str, str] | None:
36
+ """'navable_san-vit-b16_real-cur-syn.pth' -> ('san', 'real+cur+syn')."""
37
+ m = CKPT_RE.match(path)
38
+ return (m["model"], m["data"].replace("-", "+")) if m else None
39
+
40
+
41
+ def main() -> int:
42
+ p = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
43
+ p.add_argument("--real", action="store_true", help="real_ours: train / validation / test (1.5 GB)")
44
+ p.add_argument("--real-splits", nargs="+", default=nv.CONFIGS["real_ours"], choices=nv.CONFIGS["real_ours"])
45
+ p.add_argument("--curated", action="store_true", help="real_curated train (114 GB)")
46
+ p.add_argument("--syn", nargs="+", metavar="CLASS", help="syn shards holding these classes, or 'all' (816 GB)")
47
+ p.add_argument("--max-shards", type=int, help="with --syn: at most this many shards (about 1.8 GB each)")
48
+ p.add_argument("--assets", nargs="+", metavar="CLASS", help="3D assets under synthetic_objects/, or 'all' (23 GB)")
49
+ p.add_argument("--checkpoints", nargs="+", metavar="MODEL", help="released checkpoints by model, or 'all'")
50
+ p.add_argument("--train-data", nargs="+", default=TRAIN_DATA, choices=TRAIN_DATA,
51
+ help="with --checkpoints: training-data configurations to fetch (default: all four)")
52
+ p.add_argument("--local-dir", type=Path, default=Path("navable"))
53
+ p.add_argument("--dry-run", action="store_true", help="only print what would be downloaded")
54
+ args = p.parse_args()
55
+
56
+ from huggingface_hub import HfApi, snapshot_download
57
+
58
+ files = {s.rfilename: s.size or 0 for s in HfApi().dataset_info(nv.REPO_ID, files_metadata=True).siblings}
59
+ groups: dict[str, list[str]] = defaultdict(list)
60
+ groups["card and metadata"] = [f for f in ALWAYS if f in files]
61
+ if args.real:
62
+ groups["real_ours"] = sorted(f for f in files if any(f.startswith(f"real_ours/{s}/") for s in args.real_splits))
63
+ if args.curated:
64
+ groups["real_curated"] = sorted(f for f in files if f.startswith("real_curated/"))
65
+ if args.syn:
66
+ classes = None if args.syn == ["all"] else args.syn
67
+ groups["syn" + (f" ({', '.join(classes)})" if classes else "")] = nv.syn_shards(classes, args.max_shards)
68
+ if args.assets:
69
+ known = sorted({f.split("/")[1] for f in files if f.startswith("synthetic_objects/")})
70
+ wanted = known if args.assets == ["all"] else args.assets
71
+ bad = set(wanted) - set(known)
72
+ if bad:
73
+ p.error(f"no 3D assets for {sorted(bad)}; available: {known}")
74
+ groups["3D assets"] = sorted(f for f in files if f.startswith("synthetic_objects/") and f.split("/")[1] in wanted)
75
+ if args.checkpoints:
76
+ ckpts = {f: checkpoint_key(f) for f in files if checkpoint_key(f)}
77
+ models = sorted({k[0] for k in ckpts.values()})
78
+ wanted = models if args.checkpoints == ["all"] else args.checkpoints
79
+ bad = set(wanted) - set(models)
80
+ if bad:
81
+ p.error(f"unknown models {sorted(bad)}; available: {models}")
82
+ groups["checkpoints"] = sorted(f for f, (m, d) in ckpts.items() if m in wanted and d in args.train_data)
83
+ if len(groups) == 1:
84
+ p.error("choose something to download: --real, --curated, --syn, --assets or --checkpoints")
85
+
86
+ total = 0
87
+ print(f"Plan for {nv.REPO_ID} -> {args.local_dir}/")
88
+ for name, fs in groups.items():
89
+ size = sum(files[f] for f in fs)
90
+ total += size
91
+ print(f" {name:<34} {len(fs):5d} files {size / 1e9:9.2f} GB")
92
+ if name == "checkpoints":
93
+ for f in fs:
94
+ print(f" {f} ({files[f] / 1e9:.2f} GB)")
95
+ print(f" {'total':<34} {sum(map(len, groups.values())):5d} files {total / 1e9:9.2f} GB")
96
+ if args.dry_run:
97
+ return 0
98
+ allow = [f for fs in groups.values() for f in fs]
99
+ snapshot_download(nv.REPO_ID, repo_type="dataset", allow_patterns=allow, local_dir=args.local_dir)
100
+ print(f"done: {args.local_dir.resolve()}")
101
+ return 0
102
+
103
+
104
+ if __name__ == "__main__":
105
+ sys.exit(main())
tools/eval_checkpoint.py ADDED
@@ -0,0 +1,128 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ """Run a released checkpoint (navable_*.pth) on a real_ours split and report the paper's metrics.
3
+
4
+ Supports the mmsegmentation 1.x models: DeepLabv3+, EncNet, SegFormer and Mask2Former
5
+ (Mask2Former also needs mmdet). Needs torch, mmcv 2.x, mmengine and mmsegmentation 1.x, e.g.
6
+
7
+ pip install -U openmim && mim install mmengine "mmcv>=2.0.0,<2.2.0" "mmsegmentation>=1.2,<2" mmdet
8
+
9
+ SAN, SegMAN and the SAM3 head use their own code bases; see the GitHub repository linked in the
10
+ dataset card. The model config is read from the checkpoint itself, and predictions are mapped from
11
+ the checkpoint's label order to the dataset's class ids before scoring.
12
+
13
+ Examples:
14
+ python tools/download.py --checkpoints segformer --train-data real
15
+ python tools/eval_checkpoint.py --checkpoint navable/navable_segformer-real.pth --pred-dir preds_segformer
16
+ """
17
+ from __future__ import annotations
18
+
19
+ import argparse
20
+ import io
21
+ import json
22
+ import pickle
23
+ import sys
24
+ import tempfile
25
+ import time
26
+ import types
27
+ from pathlib import Path
28
+
29
+ import numpy as np
30
+ from PIL import Image
31
+
32
+ sys.path.insert(0, str(Path(__file__).resolve().parent))
33
+ import navable as nv # noqa: E402
34
+
35
+
36
+ class _Stub:
37
+ """Stands in for training-state objects (message hub, numpy scalars) that inference doesn't need."""
38
+
39
+ def __init__(self, *args, **kwargs):
40
+ pass
41
+
42
+ def __setstate__(self, state):
43
+ pass
44
+
45
+ def __call__(self, *args, **kwargs):
46
+ return _Stub()
47
+
48
+
49
+ class _Unpickler(pickle.Unpickler):
50
+ def find_class(self, module, name):
51
+ if module.split(".")[0] in ("torch", "collections", "builtins", "_codecs"):
52
+ return super().find_class(module, name)
53
+ return _Stub
54
+
55
+
56
+ def load_checkpoint(path: Path) -> tuple[dict, str]:
57
+ """(state_dict, config text) from a released checkpoint, independent of numpy/mmengine versions."""
58
+ import torch
59
+
60
+ module = types.ModuleType("navable_unpickle")
61
+ module.Unpickler, module.load = _Unpickler, pickle.load
62
+ ckpt = torch.load(path, map_location="cpu", weights_only=False, pickle_module=module)
63
+ meta = ckpt.get("meta") if isinstance(ckpt, dict) else None
64
+ cfg = meta.get("cfg") if isinstance(meta, dict) else None
65
+ if not isinstance(cfg, str) or "state_dict" not in ckpt:
66
+ raise SystemExit(f"{path.name} is not an mmsegmentation 1.x checkpoint (SAN, SegMAN and the SAM3 head "
67
+ f"need their own code; see the GitHub repository in the dataset card)")
68
+ return ckpt["state_dict"], cfg
69
+
70
+
71
+ def build_model(cfg_text: str, state_dict: dict, device: str):
72
+ from mmengine.config import Config
73
+ from mmengine.runner import load_state_dict
74
+ from mmseg.apis import init_model
75
+
76
+ with tempfile.NamedTemporaryFile("w", suffix=".py", delete=False) as f:
77
+ f.write(cfg_text)
78
+ cfg = Config.fromfile(f.name, import_custom_modules=False)
79
+ cfg.pop("custom_imports", None) # training-only plugins (dataset and metric classes)
80
+ if "SAN" in str(cfg.model.get("type")):
81
+ raise SystemExit("SAN needs its own plugin code; see the GitHub repository in the dataset card")
82
+ model = init_model(cfg, checkpoint=None, device=device)
83
+ load_state_dict(model, state_dict, strict=True)
84
+ model.dataset_meta = {"classes": nv.CHECKPOINT_CLASSES,
85
+ "palette": [nv.PALETTE[nv.CLASS_TO_ID[c]] for c in nv.CHECKPOINT_CLASSES]}
86
+ return model
87
+
88
+
89
+ def main() -> int:
90
+ p = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
91
+ p.add_argument("--checkpoint", type=Path, required=True)
92
+ p.add_argument("--split", choices=nv.CONFIGS["real_ours"], default="test")
93
+ p.add_argument("--device", default="cuda:0")
94
+ p.add_argument("--pred-dir", type=Path, help="also save predictions (dataset class ids) as PNGs")
95
+ p.add_argument("--data-dir", type=Path, help="downloaded copy of the repo (see tools/download.py)")
96
+ p.add_argument("--out", type=Path, help="write the full results as JSON")
97
+ args = p.parse_args()
98
+
99
+ from mmseg.apis import inference_model
100
+
101
+ state_dict, cfg_text = load_checkpoint(args.checkpoint)
102
+ model = build_model(cfg_text, state_dict, args.device)
103
+ if args.pred_dir:
104
+ args.pred_dir.mkdir(parents=True, exist_ok=True)
105
+
106
+ metrics, t0, n = nv.SegmentationMetrics(), time.time(), 0
107
+ for row in nv.iter_rows(nv.parquet_files("real_ours", args.split, args.data_dir), columns=["image", "mask"]):
108
+ bgr = np.ascontiguousarray(np.asarray(nv.open_image(row["image"]).convert("RGB"))[..., ::-1])
109
+ pred = inference_model(model, bgr).pred_sem_seg.data[0].cpu().numpy()
110
+ pred = nv.checkpoint_to_dataset_labels(pred)
111
+ metrics.update(pred, nv.mask_array(row["mask"]))
112
+ if args.pred_dir:
113
+ Image.fromarray(pred).save(args.pred_dir / (Path(nv.row_name(row)).stem + ".png"))
114
+ n += 1
115
+ if n % 250 == 0:
116
+ print(f" {n} images ({time.time() - t0:.0f}s)", flush=True)
117
+
118
+ r = metrics.results()
119
+ print(f"{args.checkpoint.name} on real_ours/{args.split} ({n} images): mIoU {r['mIoU']:.2f}, "
120
+ f"mPrecision {r['mPrecision']:.2f}, mRecall {r['mRecall']:.2f} (paper protocol: background + 10 classes)")
121
+ print("per-class IoU: " + ", ".join(f"{k} {v['IoU']:.1f}" for k, v in r["per_class"].items() if v["gt_pixels"]))
122
+ if args.out:
123
+ args.out.write_text(json.dumps({"checkpoint": args.checkpoint.name, "split": args.split, "images": n, **r}, indent=1))
124
+ return 0
125
+
126
+
127
+ if __name__ == "__main__":
128
+ sys.exit(main())
tools/evaluate.py ADDED
@@ -0,0 +1,87 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ """Score segmentation predictions on a real_ours split with the paper's metrics.
3
+
4
+ Predictions are label-map PNGs in --pred-dir, one per image, named like the image
5
+ (`row["image"]["path"]`, e.g. 0f4eab9c-frame_003200.png) and the same size as the image.
6
+
7
+ Reported (in %): mIoU, mPrecision and mRecall over background plus the 10 classes with real-world
8
+ examples, as in the paper's tables, the same means over the 10 classes only (fg_*), and per-class
9
+ values. revolving_door has no real-world examples and is not scored.
10
+
11
+ Examples:
12
+ python tools/evaluate.py --pred-dir preds/ # dataset class ids
13
+ python tools/evaluate.py --pred-dir preds/ --label-order checkpoint # raw output of a released mmseg checkpoint
14
+ python tools/evaluate.py --pred-dir preds/ --gt-dir data/navable/ann_dir/test
15
+ """
16
+ from __future__ import annotations
17
+
18
+ import argparse
19
+ import json
20
+ import sys
21
+ from pathlib import Path
22
+
23
+ import numpy as np
24
+ from PIL import Image
25
+
26
+ sys.path.insert(0, str(Path(__file__).resolve().parent))
27
+ import navable as nv # noqa: E402
28
+
29
+
30
+ def ground_truth(args):
31
+ """Yield (file name, label map) for every image of the split."""
32
+ if args.gt_dir:
33
+ for f in sorted(args.gt_dir.glob("*.png")):
34
+ yield f.name, nv.mask_array(Image.open(f))
35
+ return
36
+ files = nv.parquet_files("real_ours", args.split, args.data_dir)
37
+ for row in nv.iter_rows(files, columns=["image.path", "mask"]):
38
+ yield Path(nv.row_name(row)).stem + ".png", nv.mask_array(row["mask"])
39
+
40
+
41
+ def print_table(r: dict) -> None:
42
+ print(f"\n{'class':<20}{'IoU':>8}{'Prec':>8}{'Rec':>8}")
43
+ for name in nv.CLASSES:
44
+ c = r["per_class"][name]
45
+ if c["gt_pixels"] or name in ("background",):
46
+ print(f"{name:<20}{c['IoU']:8.2f}{c['precision']:8.2f}{c['recall']:8.2f}")
47
+ print(f"\n{'paper (bg + 10)':<20}{r['mIoU']:8.2f}{r['mPrecision']:8.2f}{r['mRecall']:8.2f}")
48
+ print(f"{'10 classes only':<20}{r['fg_mIoU']:8.2f}{r['fg_mPrecision']:8.2f}{r['fg_mRecall']:8.2f}")
49
+
50
+
51
+ def main() -> int:
52
+ p = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
53
+ p.add_argument("--pred-dir", type=Path, required=True)
54
+ p.add_argument("--split", choices=nv.CONFIGS["real_ours"], default="test")
55
+ p.add_argument("--label-order", choices=["dataset", "checkpoint", "checkpoint-fg"], default="dataset",
56
+ help="label ids in the predictions: dataset class ids, a released mmseg checkpoint's "
57
+ "order (12 labels), or SAN's order without background (11 labels)")
58
+ p.add_argument("--gt-dir", type=Path, help="ground-truth label PNGs (e.g. from tools/export.py) instead of the Hub")
59
+ p.add_argument("--data-dir", type=Path, help="downloaded copy of the repo (see tools/download.py)")
60
+ p.add_argument("--out", type=Path, help="write the full results as JSON")
61
+ args = p.parse_args()
62
+
63
+ metrics, missing, n = nv.SegmentationMetrics(), [], 0
64
+ for name, gt in ground_truth(args):
65
+ f = args.pred_dir / name
66
+ if not f.exists():
67
+ missing.append(name)
68
+ continue
69
+ pred = np.asarray(Image.open(f))
70
+ if args.label_order != "dataset":
71
+ pred = nv.checkpoint_to_dataset_labels(pred, has_background=args.label_order == "checkpoint")
72
+ metrics.update(pred, gt)
73
+ n += 1
74
+ if missing:
75
+ print(f"{len(missing)} images have no prediction in {args.pred_dir} (e.g. {missing[:3]})")
76
+ return 1
77
+ r = metrics.results()
78
+ print(f"real_ours/{args.split}: {n} images")
79
+ print_table(r)
80
+ if args.out:
81
+ args.out.write_text(json.dumps({"split": args.split, "images": n, **r}, indent=1))
82
+ print(f"\nresults -> {args.out}")
83
+ return 0
84
+
85
+
86
+ if __name__ == "__main__":
87
+ sys.exit(main())
tools/explore.py ADDED
@@ -0,0 +1,123 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ """Save a grid of samples: RGB, class mask, and for syn also depth, surface normals and 2D boxes.
3
+
4
+ Reads only the shards it needs from the Hub (or from a downloaded copy with --data-dir).
5
+ A syn sample costs one row group (about 170 MB); real_ours shards are about 250 MB each,
6
+ and real_curated shards about 3 GB, so download real_curated first and pass --data-dir.
7
+
8
+ Examples:
9
+ python tools/explore.py --config syn --classes escalator handrail --num 4 --out syn.png
10
+ python tools/explore.py --config real_ours --split test --num 6 --out real_test.png
11
+ """
12
+ from __future__ import annotations
13
+
14
+ import argparse
15
+ import random
16
+ import sys
17
+ from pathlib import Path
18
+
19
+ import numpy as np
20
+ from PIL import Image, ImageDraw, ImageFont
21
+
22
+ sys.path.insert(0, str(Path(__file__).resolve().parent))
23
+ import navable as nv # noqa: E402
24
+
25
+ SYN_COLUMNS = ["image", "mask", "normals", "depth", "bbox", "bbox_class_ids", "object_class", "environment", "asset"]
26
+
27
+
28
+ def font(size: int):
29
+ try:
30
+ return ImageFont.load_default(size=size)
31
+ except TypeError: # Pillow < 10.1
32
+ return ImageFont.load_default()
33
+
34
+
35
+ def panels(row: dict, config: str) -> list[tuple[str, Image.Image]]:
36
+ image, labels = nv.open_image(row["image"]).convert("RGB"), nv.mask_array(row["mask"])
37
+ out = [("RGB", image), ("class mask", nv.overlay(image, labels, alpha=0.6))]
38
+ if config == "syn":
39
+ out += [("depth (m, log scale)", nv.depth_to_rgb(nv.decode_depth(row["depth"]))),
40
+ ("surface normals", nv.normals_to_rgb(row["normals"])),
41
+ ("2D boxes", nv.draw_boxes(image, row["bbox"], row["bbox_class_ids"]))]
42
+ return out
43
+
44
+
45
+ def caption(row: dict, config: str) -> str:
46
+ if config == "syn":
47
+ return f"{row['object_class']} | {row['environment']} | {row['asset']}"
48
+ if config == "real_ours":
49
+ return f"{row['object_class']} | {row.get('location') or 'location n/a'}"
50
+ return f"{row['source_dataset']} ({row['split_origin']})"
51
+
52
+
53
+ def pick_rows(args) -> list[dict]:
54
+ """Random rows spread over classes (real_curated: over source datasets)."""
55
+ rng = random.Random(args.seed)
56
+ files = nv.parquet_files(args.config, args.split, args.data_dir, args.classes)
57
+ key = "source_dataset" if args.config == "real_curated" else "object_class"
58
+ if args.config == "syn": # shards are large: look inside `num` random shards only
59
+ files = rng.sample(files, min(args.num, len(files)))
60
+ candidates: dict[str, list[tuple[str, int]]] = {}
61
+ for f in files:
62
+ for i, k in enumerate(nv.read_column(f, key)):
63
+ if not args.classes or k in args.classes:
64
+ candidates.setdefault(k, []).append((f, i))
65
+ picks: list[tuple[str, int]] = []
66
+ pools = {k: rng.sample(v, len(v)) for k, v in sorted(candidates.items())}
67
+ while len(picks) < args.num and any(pools.values()):
68
+ for k in list(pools): # round-robin over classes
69
+ if pools[k] and len(picks) < args.num:
70
+ picks.append(pools[k].pop())
71
+ columns = SYN_COLUMNS if args.config == "syn" else None
72
+ rows = {}
73
+ for f in dict.fromkeys(f for f, _ in picks):
74
+ idx = [i for g, i in picks if g == f]
75
+ rows.update({(f, i): r for i, r in zip(idx, nv.read_rows(f, idx, columns))})
76
+ return [rows[p] for p in picks]
77
+
78
+
79
+ def grid(rows: list[dict], config: str, width: int) -> Image.Image:
80
+ cells = [panels(r, config) for r in rows]
81
+ titles = [t for t, _ in cells[0]]
82
+ heights = [max(round(width * im.height / im.width) for _, im in c) for c in cells]
83
+ head, cap, pad = 28, 22, 6
84
+ canvas = Image.new("RGB", (len(titles) * (width + pad) + pad, head + sum(h + cap + pad for h in heights)), "white")
85
+ draw = ImageDraw.Draw(canvas)
86
+ for j, t in enumerate(titles):
87
+ draw.text((pad + j * (width + pad) + 4, 6), t, fill=(30, 30, 30), font=font(16))
88
+ y = head
89
+ for row, c, h in zip(rows, cells, heights):
90
+ draw.text((pad + 4, y + 3), caption(row, config), fill=(60, 60, 60), font=font(13))
91
+ for j, (_, im) in enumerate(c):
92
+ canvas.paste(im.resize((width, round(width * im.height / im.width)), Image.BILINEAR),
93
+ (pad + j * (width + pad), y + cap))
94
+ y += h + cap + pad
95
+ return canvas
96
+
97
+
98
+ def main() -> int:
99
+ p = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
100
+ p.add_argument("--config", choices=list(nv.CONFIGS), default="real_ours")
101
+ p.add_argument("--split", default=None, help="default: test for real_ours, train otherwise")
102
+ p.add_argument("--classes", nargs="+", choices=nv.CLASSES[1:], help="syn / real_ours: only these classes")
103
+ p.add_argument("--num", type=int, default=4)
104
+ p.add_argument("--seed", type=int, default=0)
105
+ p.add_argument("--width", type=int, default=360, help="width of each panel in pixels")
106
+ p.add_argument("--data-dir", type=Path, help="downloaded copy of the repo (see tools/download.py)")
107
+ p.add_argument("--out", type=Path, default=Path("navable_samples.png"))
108
+ args = p.parse_args()
109
+ args.split = args.split or ("test" if args.config == "real_ours" else "train")
110
+ if args.split not in nv.CONFIGS[args.config]:
111
+ p.error(f"{args.config} has splits {nv.CONFIGS[args.config]}")
112
+
113
+ rows = pick_rows(args)
114
+ if not rows:
115
+ print("no matching rows")
116
+ return 1
117
+ grid(rows, args.config, args.width).save(args.out)
118
+ print(f"saved {len(rows)} samples -> {args.out}")
119
+ return 0
120
+
121
+
122
+ if __name__ == "__main__":
123
+ sys.exit(main())
tools/export.py ADDED
@@ -0,0 +1,87 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ """Export a config/split to image and label folders (the mmsegmentation layout), for training code
3
+ that reads files instead of `datasets`.
4
+
5
+ <out>/img_dir/<split>/<name>.png images, written as stored (syn renders are RGBA)
6
+ <out>/ann_dir/<split>/<name>.png label maps: pixel value = class id
7
+ <out>/depth_dir/<split>/<name>.npy syn only, with --with-depth (float16 metres)
8
+ <out>/normal_dir/<split>/<name>.png syn only, with --with-normals
9
+ <out>/classes.json class names and colors in the label order used
10
+
11
+ --label-order checkpoint writes labels in the order the released checkpoints were trained with
12
+ (see navable.CHECKPOINT_CLASSES), so their mmseg configs can be reused unchanged.
13
+
14
+ Examples:
15
+ python tools/export.py --config real_ours --split train --out data/navable
16
+ python tools/export.py --config real_ours --split test --out data/navable
17
+ python tools/export.py --config syn --classes escalator --max-rows 2000 --with-depth --out data/navable_syn
18
+ """
19
+ from __future__ import annotations
20
+
21
+ import argparse
22
+ import io
23
+ import json
24
+ import sys
25
+ from pathlib import Path
26
+
27
+ import numpy as np
28
+ from PIL import Image
29
+
30
+ sys.path.insert(0, str(Path(__file__).resolve().parent))
31
+ import navable as nv # noqa: E402
32
+
33
+
34
+ def main() -> int:
35
+ p = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
36
+ p.add_argument("--config", choices=list(nv.CONFIGS), required=True)
37
+ p.add_argument("--split", default="train")
38
+ p.add_argument("--out", type=Path, required=True)
39
+ p.add_argument("--classes", nargs="+", choices=nv.CLASSES[1:], help="syn / real_ours: only rows of these classes")
40
+ p.add_argument("--max-rows", type=int, help="stop after this many rows")
41
+ p.add_argument("--label-order", choices=["dataset", "checkpoint"], default="dataset")
42
+ p.add_argument("--with-depth", action="store_true", help="syn: also write depth (.npy, float16, metres)")
43
+ p.add_argument("--with-normals", action="store_true", help="syn: also write surface normals (.png)")
44
+ p.add_argument("--data-dir", type=Path, help="downloaded copy of the repo (see tools/download.py)")
45
+ args = p.parse_args()
46
+ if args.split not in nv.CONFIGS[args.config]:
47
+ p.error(f"{args.config} has splits {nv.CONFIGS[args.config]}")
48
+ if (args.with_depth or args.with_normals) and args.config != "syn":
49
+ p.error("--with-depth / --with-normals only apply to syn")
50
+ if args.classes and args.config == "real_curated":
51
+ p.error("real_curated rows have no object_class; export it whole")
52
+
53
+ columns = ["image", "mask"] + (["depth"] if args.with_depth else []) + (["normals"] if args.with_normals else [])
54
+ dirs = {"img": "img_dir", "ann": "ann_dir", "depth": "depth_dir", "normals": "normal_dir"}
55
+ for key in ["img", "ann"] + (["depth"] if args.with_depth else []) + (["normals"] if args.with_normals else []):
56
+ (args.out / dirs[key] / args.split).mkdir(parents=True, exist_ok=True)
57
+
58
+ order = nv.CHECKPOINT_CLASSES if args.label_order == "checkpoint" else nv.CLASSES
59
+ (args.out / "classes.json").write_text(json.dumps(
60
+ {"label_order": args.label_order, "classes": order,
61
+ "palette": [nv.PALETTE[nv.CLASS_TO_ID[c]] for c in order], "ignore_index": nv.IGNORE_INDEX}, indent=1))
62
+
63
+ files = nv.parquet_files(args.config, args.split, args.data_dir, args.classes)
64
+ n = 0
65
+ for row in nv.iter_rows(files, columns, args.classes):
66
+ stem = Path(nv.row_name(row)).stem
67
+ (args.out / "img_dir" / args.split / f"{stem}.png").write_bytes(row["image"]["bytes"])
68
+ ann = args.out / "ann_dir" / args.split / f"{stem}.png"
69
+ if args.label_order == "dataset":
70
+ ann.write_bytes(row["mask"]["bytes"]) # palettized PNG; pixel value = dataset class id
71
+ else:
72
+ Image.fromarray(nv.dataset_to_checkpoint_labels(nv.mask_array(row["mask"]))).save(ann)
73
+ if args.with_depth:
74
+ (args.out / "depth_dir" / args.split / f"{stem}.npy").write_bytes(row["depth"])
75
+ if args.with_normals:
76
+ (args.out / "normal_dir" / args.split / f"{stem}.png").write_bytes(row["normals"]["bytes"])
77
+ n += 1
78
+ if n % 500 == 0:
79
+ print(f" {n} rows", flush=True)
80
+ if args.max_rows and n >= args.max_rows:
81
+ break
82
+ print(f"exported {n} rows of {args.config}/{args.split} -> {args.out}")
83
+ return 0
84
+
85
+
86
+ if __name__ == "__main__":
87
+ sys.exit(main())
tools/navable.py ADDED
@@ -0,0 +1,367 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Helpers for the NavAble dataset: https://huggingface.co/datasets/NavAble/NeurIPS_2026_BLV
2
+
3
+ Get the tools folder (this file plus the command-line scripts next to it) with
4
+
5
+ hf download NavAble/NeurIPS_2026_BLV --repo-type dataset --include "tools/*" --local-dir navable
6
+
7
+ then `import navable` from `navable/tools`. The module only needs numpy and Pillow;
8
+ loading data needs `datasets`, and `NavAbleSegmentation` needs torch.
9
+ """
10
+ from __future__ import annotations
11
+
12
+ import io
13
+ import json
14
+ from pathlib import Path
15
+
16
+ import numpy as np
17
+ from PIL import Image, ImageDraw
18
+
19
+ REPO_ID = "NavAble/NeurIPS_2026_BLV"
20
+ CONFIGS = {"syn": ["train"], "real_ours": ["train", "validation", "test"], "real_curated": ["train"]}
21
+
22
+ # Mask pixel value -> class name and color (same as class_index.json).
23
+ CLASSES = ["background", "aps_button", "bus_stop", "bus_stop_sign", "crosswalk", "door_button",
24
+ "elevator", "elevator_button", "escalator", "handrail", "pedestrian_signal", "revolving_door"]
25
+ PALETTE = [(0, 0, 0), (220, 20, 60), (255, 140, 0), (255, 215, 0), (50, 205, 50), (0, 191, 255),
26
+ (138, 43, 226), (255, 105, 180), (0, 128, 128), (165, 42, 42), (75, 0, 130), (255, 20, 147)]
27
+ CLASS_TO_ID = {name: i for i, name in enumerate(CLASSES)}
28
+ IGNORE_INDEX = 255
29
+
30
+ # Real-world metrics use the 10 classes that have real examples; revolving_door is synthetic-only.
31
+ # The paper's mIoU averages over background plus these 10 classes.
32
+ EVAL_CLASS_IDS = list(range(1, 11))
33
+ PAPER_CLASS_IDS = [0] + EVAL_CLASS_IDS
34
+ SYNTHETIC_ONLY_IDS = [11]
35
+
36
+ # The released checkpoints (navable_*.pth) were trained with a different label order.
37
+ # mmseg models (DeepLabv3+, EncNet, SegFormer, Mask2Former) and SegMAN predict 12 labels in
38
+ # this order; SAN predicts the 11 foreground labels in the same order without background.
39
+ CHECKPOINT_CLASSES = ["background", "elevator", "elevator_button", "door_button", "crosswalk",
40
+ "pedestrian_signal", "aps_button", "bus_stop", "bus_stop_sign", "handrail",
41
+ "escalator", "revolving_door"]
42
+ _CKPT_TO_DATASET = np.array([CLASS_TO_ID[c] for c in CHECKPOINT_CLASSES], dtype=np.uint8)
43
+ _DATASET_TO_CKPT = np.argsort(_CKPT_TO_DATASET).astype(np.uint8)
44
+
45
+
46
+ # --------------------------------------------------------------------------- loading
47
+
48
+ def load(config: str, split: str, streaming: bool = False, **kwargs):
49
+ """`datasets.load_dataset` for one config and split. Use streaming=True to avoid a full download."""
50
+ from datasets import load_dataset
51
+
52
+ if split not in CONFIGS.get(config, []):
53
+ raise ValueError(f"unknown config/split {config}/{split}; choose from {CONFIGS}")
54
+ return load_dataset(REPO_ID, config, split=split, streaming=streaming, **kwargs)
55
+
56
+
57
+ def syn_shard_index(local_dir: str | Path | None = None) -> dict:
58
+ """Contents of each syn/train shard (rows per class, environments, size), from syn/shard_index.json."""
59
+ if local_dir and (Path(local_dir) / "syn/shard_index.json").exists():
60
+ path = Path(local_dir) / "syn/shard_index.json"
61
+ else:
62
+ from huggingface_hub import hf_hub_download
63
+ path = hf_hub_download(REPO_ID, "syn/shard_index.json", repo_type="dataset")
64
+ return json.loads(Path(path).read_text())
65
+
66
+
67
+ def syn_shards(classes: list[str] | None = None, max_shards: int | None = None,
68
+ local_dir: str | Path | None = None) -> list[str]:
69
+ """Repo paths of the syn/train shards that contain any of `classes` (all shards if None)."""
70
+ unknown = set(classes or []) - set(CLASSES[1:])
71
+ if unknown:
72
+ raise ValueError(f"unknown classes {sorted(unknown)}; choose from {CLASSES[1:]}")
73
+ shards = [s["file"] for s in syn_shard_index(local_dir)["shards"]
74
+ if classes is None or set(s["classes"]) & set(classes)]
75
+ return shards[:max_shards] if max_shards else shards
76
+
77
+
78
+ def load_syn(classes: list[str] | None = None, max_shards: int | None = None, streaming: bool = True, **kwargs):
79
+ """Load only the syn rows of `classes`, reading just the shards that hold them.
80
+
81
+ For example `load_syn(["escalator"])` reads 8 of the 451 shards (about 12 GB) instead of 816 GB.
82
+ `max_shards` caps the shards read per class, for a quick sample.
83
+ """
84
+ from datasets import concatenate_datasets, load_dataset
85
+
86
+ def parquet(shards):
87
+ files = [f"hf://datasets/{REPO_ID}/{s['file']}" for s in shards]
88
+ return load_dataset("parquet", data_files={"train": files}, split="train", streaming=streaming, **kwargs)
89
+
90
+ index = syn_shard_index()["shards"]
91
+ if not classes:
92
+ return parquet(index[:max_shards] if max_shards else index)
93
+ unknown = set(classes) - set(CLASSES[1:])
94
+ if unknown:
95
+ raise ValueError(f"unknown classes {sorted(unknown)}; choose from {CLASSES[1:]}")
96
+ # Rows are sorted by class (in class-id order) across and within shards, so each class is one
97
+ # contiguous run: skip the previous class's rows in its first shard, take its own rows.
98
+ parts = []
99
+ for cls in sorted(set(classes), key=CLASS_TO_ID.get):
100
+ shards = [s for s in index if cls in s["classes"]][:max_shards]
101
+ lead = sum(n for c, n in shards[0]["classes"].items() if CLASS_TO_ID[c] < CLASS_TO_ID[cls])
102
+ parts.append(parquet(shards).skip(lead).take(sum(s["classes"][cls] for s in shards)))
103
+ return parts[0] if len(parts) == 1 else concatenate_datasets(parts)
104
+
105
+
106
+ def parquet_files(config: str, split: str, data_dir: str | Path | None = None,
107
+ classes: list[str] | None = None, max_shards: int | None = None) -> list[str]:
108
+ """Parquet shards of a config/split: local paths under `data_dir` (a downloaded copy of the
109
+ repo) or hf:// paths read on demand. For syn, `classes` keeps only the shards holding them."""
110
+ if config == "syn":
111
+ rel = syn_shards(classes, max_shards, local_dir=data_dir)
112
+ else:
113
+ from huggingface_hub import HfFileSystem
114
+ pattern = f"{config}/{split}/data-*.parquet"
115
+ if data_dir:
116
+ rel = sorted(str(p.relative_to(data_dir)) for p in Path(data_dir).glob(pattern))
117
+ else:
118
+ rel = sorted(p.split(f"{REPO_ID}/", 1)[1] for p in HfFileSystem().glob(f"datasets/{REPO_ID}/{pattern}"))
119
+ rel = rel[:max_shards] if max_shards else rel
120
+ if data_dir:
121
+ missing = [r for r in rel if not (Path(data_dir) / r).exists()]
122
+ if missing:
123
+ raise FileNotFoundError(f"{len(missing)} shards missing under {data_dir} (e.g. {missing[0]}); "
124
+ f"download them with tools/download.py or drop --data-dir")
125
+ return [str(Path(data_dir) / r) for r in rel]
126
+ return [f"hf://datasets/{REPO_ID}/{r}" for r in rel]
127
+
128
+
129
+ def _open(path: str):
130
+ if path.startswith("hf://"):
131
+ from huggingface_hub import HfFileSystem
132
+ return HfFileSystem().open(path, "rb", block_size=8 << 20)
133
+ return open(path, "rb")
134
+
135
+
136
+ def iter_rows(files: list[str], columns: list[str] | None = None, classes: list[str] | None = None,
137
+ batch_size: int = 16):
138
+ """Yield rows (dicts) from parquet shards without `datasets`.
139
+
140
+ Image columns come back as {"bytes", "path"} dicts; `open_image` decodes them. With `classes`,
141
+ rows of other classes are skipped (syn shards at class boundaries hold two classes).
142
+ """
143
+ import pyarrow.parquet as pq
144
+
145
+ wanted = set(classes) if classes else None
146
+ cols = columns if not (wanted and columns and "object_class" not in columns) else columns + ["object_class"]
147
+ for f in files:
148
+ with _open(f) as fh:
149
+ for batch in pq.ParquetFile(fh).iter_batches(batch_size=batch_size, columns=cols):
150
+ for row in batch.to_pylist():
151
+ if wanted is None or row.get("object_class") in wanted:
152
+ yield row
153
+
154
+
155
+ def read_column(file: str, column: str) -> list:
156
+ """One small column of a shard (e.g. object_class), without reading the images."""
157
+ import pyarrow.parquet as pq
158
+
159
+ with _open(file) as fh:
160
+ return pq.ParquetFile(fh).read(columns=[column]).column(0).to_pylist()
161
+
162
+
163
+ def read_rows(file: str, indices: list[int], columns: list[str] | None = None) -> list[dict]:
164
+ """Rows at `indices` of one shard, reading only the row groups that contain them."""
165
+ import pyarrow.parquet as pq
166
+
167
+ with _open(file) as fh:
168
+ pf = pq.ParquetFile(fh)
169
+ starts = np.cumsum([0] + [pf.metadata.row_group(g).num_rows for g in range(pf.num_row_groups)])
170
+ by_group: dict[int, list[int]] = {}
171
+ for i in indices:
172
+ g = int(np.searchsorted(starts, i, side="right") - 1)
173
+ by_group.setdefault(g, []).append(i)
174
+ rows = {}
175
+ for g, idx in by_group.items():
176
+ # Batches, not read_row_group: real_curated row groups are too large to read in one piece.
177
+ offset, want, last = int(starts[g]), set(idx), max(idx)
178
+ for batch in pf.iter_batches(batch_size=16, row_groups=[g], columns=columns):
179
+ for j, row in enumerate(batch.to_pylist()):
180
+ if offset + j in want:
181
+ rows[offset + j] = row
182
+ offset += batch.num_rows
183
+ if offset > last:
184
+ break
185
+ return [rows[i] for i in indices]
186
+
187
+
188
+ def open_image(value) -> Image.Image:
189
+ """Decode an image cell: a {"bytes", "path"} dict from iter_rows, or a PIL image from `datasets`."""
190
+ return Image.open(io.BytesIO(value["bytes"])) if isinstance(value, dict) else value
191
+
192
+
193
+ def row_name(row) -> str:
194
+ """Stable per-row file name (unique within each config and split), e.g. '0f4eab9c-frame_003200.png'."""
195
+ img = row["image"]
196
+ return img["path"] if isinstance(img, dict) else img.filename
197
+
198
+
199
+ # --------------------------------------------------------------------------- decoding
200
+
201
+ def mask_array(mask) -> np.ndarray:
202
+ """Label map (H, W) uint8 from a mask cell, PIL image or array; values are class ids (see CLASSES)."""
203
+ return np.asarray(open_image(mask) if isinstance(mask, dict) else mask, dtype=np.uint8)
204
+
205
+
206
+ def decode_depth(value) -> np.ndarray:
207
+ """syn `depth` column -> float16 array (720, 1280): distance to the image plane in metres, inf where nothing was hit."""
208
+ return np.load(io.BytesIO(value))
209
+
210
+
211
+ def checkpoint_to_dataset_labels(pred: np.ndarray, has_background: bool = True) -> np.ndarray:
212
+ """Map a released checkpoint's predicted label map to the dataset's class ids.
213
+
214
+ has_background=False is for SAN, whose label 0 is elevator (CHECKPOINT_CLASSES[1]).
215
+ """
216
+ pred = np.asarray(pred).astype(np.int64)
217
+ return _CKPT_TO_DATASET[pred if has_background else pred + 1]
218
+
219
+
220
+ def dataset_to_checkpoint_labels(mask: np.ndarray) -> np.ndarray:
221
+ """Map dataset class ids to the checkpoints' training label order (255 stays ignored)."""
222
+ mask = np.asarray(mask)
223
+ out = _DATASET_TO_CKPT[np.minimum(mask, len(CLASSES) - 1)]
224
+ return np.where(mask == IGNORE_INDEX, IGNORE_INDEX, out).astype(np.uint8)
225
+
226
+
227
+ # --------------------------------------------------------------------------- visualization
228
+
229
+ def colorize(labels: np.ndarray) -> Image.Image:
230
+ """Color a label map with the dataset palette (ignored or unknown labels are white)."""
231
+ lut = np.full((256, 3), 255, dtype=np.uint8)
232
+ lut[: len(PALETTE)] = PALETTE
233
+ return Image.fromarray(lut[mask_array(labels)])
234
+
235
+
236
+ def overlay(image, labels, alpha: float = 0.5) -> Image.Image:
237
+ """Blend the colored mask over the image; background pixels are left untouched."""
238
+ image = open_image(image) if isinstance(image, dict) else image
239
+ img = np.asarray(Image.fromarray(np.asarray(image)).convert("RGB"), dtype=np.float32)
240
+ lab = mask_array(labels)
241
+ col = np.asarray(colorize(lab), dtype=np.float32)
242
+ fg = (lab > 0)[..., None]
243
+ return Image.fromarray(np.where(fg, (1 - alpha) * img + alpha * col, img).astype(np.uint8))
244
+
245
+
246
+ _TURBO = np.array([[48, 18, 59], [70, 107, 227], [41, 187, 236], [49, 242, 153], [164, 252, 60],
247
+ [237, 208, 58], [251, 128, 34], [207, 46, 3], [122, 4, 3]], dtype=np.float32)
248
+
249
+
250
+ def depth_to_rgb(depth: np.ndarray, max_depth: float | None = None) -> Image.Image:
251
+ """Color a depth map (near = blue, far = red) on a log scale; missing depth (inf) is black."""
252
+ d = np.asarray(depth, dtype=np.float32)
253
+ valid = np.isfinite(d) & (d > 0)
254
+ if not valid.any():
255
+ return Image.fromarray(np.zeros(d.shape + (3,), np.uint8))
256
+ lo = float(np.percentile(d[valid], 1))
257
+ hi = float(max_depth or np.percentile(d[valid], 99))
258
+ t = np.clip((np.log(np.clip(d, lo, hi)) - np.log(lo)) / max(np.log(hi) - np.log(lo), 1e-6), 0, 1)
259
+ x = t * (len(_TURBO) - 1)
260
+ i = np.clip(x.astype(int), 0, len(_TURBO) - 2)
261
+ rgb = _TURBO[i] + (x - i)[..., None] * (_TURBO[i + 1] - _TURBO[i])
262
+ rgb[~valid] = 0
263
+ return Image.fromarray(rgb.astype(np.uint8))
264
+
265
+
266
+ def normals_to_rgb(normals) -> Image.Image:
267
+ """syn `normals` image (RGBA as rendered) -> RGB for display."""
268
+ normals = open_image(normals) if isinstance(normals, dict) else normals
269
+ return normals.convert("RGB") if isinstance(normals, Image.Image) else Image.fromarray(np.asarray(normals)[..., :3])
270
+
271
+
272
+ def draw_boxes(image, boxes, class_ids, width: int = 3, labels: bool = True) -> Image.Image:
273
+ """Draw 2D boxes [x_min, y_min, x_max, y_max] (pixels) in their class colors."""
274
+ image = open_image(image) if isinstance(image, dict) else image
275
+ img = Image.fromarray(np.asarray(image)).convert("RGB")
276
+ draw = ImageDraw.Draw(img)
277
+ for box, cid in zip(boxes, class_ids):
278
+ color = PALETTE[cid] if 0 <= cid < len(PALETTE) else (255, 255, 255)
279
+ draw.rectangle([float(v) for v in box], outline=color, width=width)
280
+ if labels:
281
+ name = CLASSES[cid] if 0 <= cid < len(CLASSES) else str(cid)
282
+ x, y = float(box[0]), max(float(box[1]) - 12, 0)
283
+ draw.rectangle(draw.textbbox((x, y), name), fill=color)
284
+ draw.text((x, y), name, fill=(0, 0, 0) if sum(color) > 380 else (255, 255, 255))
285
+ return img
286
+
287
+
288
+ # --------------------------------------------------------------------------- metrics
289
+
290
+ class SegmentationMetrics:
291
+ """Confusion-matrix IoU / precision / recall, following the paper's protocol.
292
+
293
+ The paper's mIoU, mPrecision and mRecall average over background plus the 10 classes that
294
+ have real-world examples (PAPER_CLASS_IDS); `fg_*` averages over the 10 classes only. Pixels
295
+ labelled IGNORE_INDEX in the ground truth are skipped.
296
+ """
297
+
298
+ def __init__(self, num_classes: int = len(CLASSES), ignore_index: int = IGNORE_INDEX):
299
+ self.num_classes = num_classes
300
+ self.ignore_index = ignore_index
301
+ self.confusion = np.zeros((num_classes, num_classes), dtype=np.int64) # rows: ground truth
302
+
303
+ def update(self, pred, gt) -> None:
304
+ pred, gt = np.asarray(pred).astype(np.int64), np.asarray(gt).astype(np.int64)
305
+ if pred.shape != gt.shape:
306
+ raise ValueError(f"prediction {pred.shape} and ground truth {gt.shape} differ in size")
307
+ keep = gt != self.ignore_index
308
+ pred, gt = np.clip(pred[keep], 0, self.num_classes - 1), gt[keep]
309
+ self.confusion += np.bincount(gt * self.num_classes + pred,
310
+ minlength=self.num_classes ** 2).reshape(self.num_classes, self.num_classes)
311
+
312
+ def results(self, class_ids: list[int] | None = None) -> dict:
313
+ """Mean and per-class IoU, precision and recall in % (means over `class_ids`, default PAPER_CLASS_IDS)."""
314
+ class_ids = PAPER_CLASS_IDS if class_ids is None else class_ids
315
+ tp = np.diag(self.confusion).astype(np.float64)
316
+ gt, pred = self.confusion.sum(1), self.confusion.sum(0)
317
+ with np.errstate(divide="ignore", invalid="ignore"):
318
+ iou, prec, rec = tp / (gt + pred - tp), tp / pred, tp / gt
319
+ per_class = {CLASSES[c]: {"IoU": float(100 * iou[c]), "precision": float(100 * prec[c]),
320
+ "recall": float(100 * rec[c]), "gt_pixels": int(gt[c])}
321
+ for c in range(self.num_classes)}
322
+
323
+ def means(ids):
324
+ ids = [c for c in ids if gt[c] > 0] # classes absent from the ground truth are skipped
325
+ with np.errstate(invalid="ignore"):
326
+ return [float(100 * np.nanmean(a[ids])) if ids else float("nan") for a in (iou, prec, rec)], ids
327
+
328
+ (miou, mprec, mrec), scored = means(class_ids)
329
+ (fg_iou, fg_prec, fg_rec), _ = means([c for c in class_ids if c != 0])
330
+ return {"mIoU": miou, "mPrecision": mprec, "mRecall": mrec,
331
+ "fg_mIoU": fg_iou, "fg_mPrecision": fg_prec, "fg_mRecall": fg_rec,
332
+ "classes_scored": [CLASSES[c] for c in scored], "per_class": per_class}
333
+
334
+
335
+ # --------------------------------------------------------------------------- PyTorch
336
+
337
+ def NavAbleSegmentation(hf_dataset, size: int | tuple[int, int] | None = None, with_depth: bool = False):
338
+ """Wrap a (non-streaming) `datasets` split as a torch Dataset of {"image", "mask"[, "depth"]} tensors.
339
+
340
+ image: float32 (3, H, W) in [0, 1]; mask: int64 (H, W) class ids; depth: float32 (1, H, W) metres,
341
+ syn only. `size` resizes (bilinear for images, nearest for masks and depth).
342
+ """
343
+ import torch
344
+ from torch.utils.data import Dataset
345
+
346
+ hw = (size, size) if isinstance(size, int) else size
347
+
348
+ class _Seg(Dataset):
349
+ def __len__(self):
350
+ return len(hf_dataset)
351
+
352
+ def __getitem__(self, idx):
353
+ row = hf_dataset[idx]
354
+ img, mask = row["image"].convert("RGB"), row["mask"]
355
+ if hw:
356
+ img = img.resize(hw[::-1], Image.BILINEAR)
357
+ mask = mask.resize(hw[::-1], Image.NEAREST)
358
+ out = {"image": torch.from_numpy(np.asarray(img, dtype=np.float32) / 255).permute(2, 0, 1),
359
+ "mask": torch.from_numpy(mask_array(mask).astype(np.int64))}
360
+ if with_depth:
361
+ depth = decode_depth(row["depth"]).astype(np.float32)
362
+ if hw:
363
+ depth = np.asarray(Image.fromarray(depth).resize(hw[::-1], Image.NEAREST))
364
+ out["depth"] = torch.from_numpy(depth)[None]
365
+ return out
366
+
367
+ return _Seg()
tools/requirements.txt ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ # Core: loading, exploring, exporting and evaluating
2
+ datasets>=2.19
3
+ huggingface_hub>=0.23
4
+ pyarrow>=14
5
+ numpy
6
+ pillow>=9.1
7
+ # Optional: NavAbleSegmentation (PyTorch) and eval_checkpoint.py
8
+ # torch
9
+ # mmengine, mmcv>=2.0.0,<2.2.0, mmsegmentation>=1.2,<2, mmdet (Mask2Former)