hljiang commited on
Commit
33781e6
·
verified ·
1 Parent(s): 9c9aa60

Restyle dataset card (MoGe-like table, download, references from paper)

Browse files
Files changed (1) hide show
  1. README.md +53 -96
README.md CHANGED
@@ -1,5 +1,5 @@
1
  ---
2
- license: other
3
  task_categories:
4
  - depth-estimation
5
  language:
@@ -11,156 +11,113 @@ tags:
11
  - depth
12
  - da360
13
  - benchmark
 
14
  pretty_name: DA360 Benchmark Test Sets
15
  size_categories:
16
  - 1K<n<10K
17
- # Evaluation-only: every config exposes a single `test` split (no train/val).
18
  configs:
19
  - config_name: metropolis
20
- description: Metropolis outdoor ERP test benchmark (archive).
21
  data_files:
22
  - split: test
23
  path: archives/Metropolis.zip
24
  - config_name: matterport3d
25
- description: Matterport3D indoor ERP test benchmark (archive).
26
  data_files:
27
  - split: test
28
  path: archives/Matterport3D.zip
29
  - config_name: stanford2d3d
30
- description: Stanford2D3D indoor ERP test benchmark (area_5a/5b, archive).
31
  data_files:
32
  - split: test
33
  path: archives/Stanford2D3D.zip
34
  ---
35
 
36
- # DA360 Benchmark Test Sets
37
 
38
- Evaluation data for **[DA360: Depth Anything in 360°](https://antigravity-tech.github.io/DA360/)** ([paper](https://arxiv.org/abs/2512.22819) · [code](https://github.com/antigravity-tech/DA360) · [demo Space](https://huggingface.co/spaces/hljiang/DA360)).
39
 
40
- **All subsets are test-only evaluation data** (no training split). In the Dataset Viewer, pick a config (`metropolis`, `matterport3d`, `stanford2d3d`) and split **`test`**.
41
 
42
- Three ERP (equirectangular) test benchmarks from the paper:
 
 
 
 
43
 
44
- | Benchmark | Setting | Role in paper | Scenes / areas |
45
- |-----------|---------|---------------|----------------|
46
- | **Metropolis** | Outdoor | Primary large-scale ERP benchmark | RGB + depth pairs under `rgb/`, `depth/` |
47
- | **Matterport3D** | Indoor | Standard indoor ERP test | 4 scenes, `pano_skybox_color/` + `pano_depth/` |
48
- | **Stanford2D3D** | Indoor | Standard indoor ERP test | `area_5a`, `area_5b` — `rgb_/`, `depth/` |
49
 
50
  ---
51
 
52
- ## Start here (Files tab)
53
 
54
- All benchmarks are distributed as **archives only** (keeps the Files tab simple):
55
 
 
 
56
  ```
57
- hljiang/DA360/
58
- ├── README.md ← dataset card (this page)
59
- ├── scripts/prepare_data.sh ← recommended: build ./data/ for official eval
60
- └── archives/ ← one .zip per benchmark
61
- ├── Metropolis.zip
62
- ├── Stanford2D3D.zip
63
- └── Matterport3D.zip
64
- ```
65
-
66
- | If you want… | Go to… |
67
- |--------------|--------|
68
- | All three benchmarks | [`archives/`](archives/) or [`scripts/prepare_data.sh`](scripts/prepare_data.sh) |
69
- | A single split | [`archives/Metropolis.zip`](archives/Metropolis.zip), [`archives/Matterport3D.zip`](archives/Matterport3D.zip), [`archives/Stanford2D3D.zip`](archives/Stanford2D3D.zip) |
70
 
71
- ---
72
-
73
- ## Local layout (official DA360 eval code)
74
-
75
- Clone [antigravity-tech/DA360](https://github.com/antigravity-tech/DA360) and point `./data/` at this dataset:
76
 
 
 
77
  ```
78
- data/
79
- ├── Metropolis/
80
- │ ├── rgb/*.jpg
81
- │ └── depth/*.png
82
- ├── Matterport3D/
83
- │ └── {scene_id}/pano_skybox_color|pano_depth/
84
- └── Stanford2D3D/
85
- └── {area_5a|area_5b}/rgb_|depth/
86
- ```
87
-
88
- Then run evaluation, e.g. `bash scripts/evaluate.sh`.
89
 
90
- ### Recommended download (archives → `data/`)
91
 
92
  ```bash
93
- git clone https://github.com/antigravity-tech/DA360
94
- cd DA360
95
- bash -c "$(curl -fsSL https://huggingface.co/datasets/hljiang/DA360/raw/main/scripts/prepare_data.sh)" # or copy scripts/prepare_data.sh from this repo
96
- # default output: ./data
97
  ```
98
 
99
- Or with Python:
100
 
101
  ```bash
102
- pip install huggingface_hub
103
- python - <<'PY'
104
- from pathlib import Path
105
- from huggingface_hub import hf_hub_download
106
- import zipfile
107
-
108
- repo = "hljiang/DA360"
109
- out = Path("data")
110
- out.mkdir(parents=True, exist_ok=True)
111
- for name in ("Metropolis", "Stanford2D3D", "Matterport3D"):
112
- z = Path(hf_hub_download(repo_id=repo, repo_type="dataset", filename=f"archives/{name}.zip"))
113
- with zipfile.ZipFile(z) as zf:
114
- zf.extractall(out)
115
- print("Done. Expect:", out.resolve())
116
- PY
117
  ```
118
 
119
- ### Single archive (one benchmark)
120
 
121
- ```python
122
- from huggingface_hub import hf_hub_download
123
- import zipfile
124
- from pathlib import Path
125
 
126
- repo = "hljiang/DA360"
127
- name = "Matterport3D" # or Metropolis, Stanford2D3D
128
- out = Path("data")
129
- out.mkdir(parents=True, exist_ok=True)
130
- z = Path(hf_hub_download(repo_id=repo, repo_type="dataset", filename=f"archives/{name}.zip"))
131
- with zipfile.ZipFile(z) as zf:
132
- zf.extractall(out)
133
  ```
134
 
135
  ---
136
 
137
- ## Depth formats
 
 
 
 
 
 
138
 
139
- | Benchmark | Depth file | Meters |
140
- |-----------|------------|--------|
141
- | Metropolis | 16-bit PNG | `pixel / 512` |
142
- | Matterport3D | 16-bit PNG | `pixel / 4000` |
143
- | Stanford2D3D | 16-bit PNG | `pixel / 512` |
144
 
145
  ---
146
 
147
- ## Matterport3D scenes (inside `archives/Matterport3D.zip`)
148
 
149
- | Scene ID | Layout |
150
- |----------|--------|
151
- | `2t7WUuJeko7` | `pano_skybox_color/`, `pano_depth/` |
152
- | `5ZKStnWn8Zo` | same |
153
- | `ARNzJeq3xxb` | same |
154
- | `fzynW3qQPVF` | same |
155
 
156
- ---
157
 
158
- ## Citation
159
 
160
- If you use these benchmarks, please cite the DA360 paper (see [project page](https://antigravity-tech.github.io/DA360/)).
161
 
162
  ---
163
 
164
- ## Status
165
 
166
- Dataset repository for paper reproduction. Visibility may be **private** during review; links above remain valid for authorized users.
 
1
  ---
2
+ license: apache-2.0
3
  task_categories:
4
  - depth-estimation
5
  language:
 
11
  - depth
12
  - da360
13
  - benchmark
14
+ - geometry
15
  pretty_name: DA360 Benchmark Test Sets
16
  size_categories:
17
  - 1K<n<10K
18
+ arxiv: 2512.22819
19
  configs:
20
  - config_name: metropolis
21
+ description: Metropolis outdoor ERP test benchmark.
22
  data_files:
23
  - split: test
24
  path: archives/Metropolis.zip
25
  - config_name: matterport3d
26
+ description: Matterport3D indoor ERP test benchmark.
27
  data_files:
28
  - split: test
29
  path: archives/Matterport3D.zip
30
  - config_name: stanford2d3d
31
+ description: Stanford2D3D indoor ERP test benchmark (Area 5).
32
  data_files:
33
  - split: test
34
  path: archives/Stanford2D3D.zip
35
  ---
36
 
37
+ # DA360 benchmark test sets
38
 
39
+ Processed **equirectangular (ERP) RGB–depth** test splits used to evaluate [DA360: Depth Anything in 360°](https://antigravity-tech.github.io/DA360/) ([arXiv:2512.22819](https://arxiv.org/abs/2512.22819) · [code](https://github.com/antigravity-tech/DA360) · [demo](https://huggingface.co/spaces/hljiang/DA360)).
40
 
41
+ All subsets are **test-only** (no train/validation). Each benchmark is shipped as one zip under [`archives/`](archives/).
42
 
43
+ | Dataset | Source | Publication | Num images | Storage size | Note |
44
+ | ------- | ------ | ----------- | ---------- | ------------ | ---- |
45
+ | **Metropolis** | [Mapillary Metropolis](https://www.mapillary.com/dataset/metropolis) | [DA360 Sec. 3.3](https://arxiv.org/abs/2512.22819) | 3,000 | 3.1 GB | **New outdoor ERP benchmark.** Curated from ~10k synchronized Mapillary samples; quality-filtered; bottom ERP inpainted (UniFuse-style); cubemap depth reprojected to ERP. 16-bit depth, meters = pixel / 512. |
46
+ | **Matterport3D** | [Matterport3D](https://niessner.github.io/Matterport/) | [1] | 2,014 | 2.8 GB | Standard **18-scene indoor test split** (same protocol as prior panoramic works). `pano_skybox_color/` + `pano_depth/`. 16-bit depth, meters = pixel / 4000. |
47
+ | **Stanford2D3D** | [Stanford 2D-3D-S](http://buildingparser.stanford.edu/dataset.html) | [2] | 373 | 0.9 GB | Standard **Area 5** test set (`area_5a`, `area_5b`). 16-bit depth, meters = pixel / 512. |
48
 
49
+ **Total (all zips):** ~6.8 GB · **~6.4k** ERP pairs combined.
50
+
51
+ Evaluation in the paper uses **518×1036** ERP resolution. See the official repo for metrics and `bash scripts/evaluate.sh`.
 
 
52
 
53
  ---
54
 
55
+ ## Download
56
 
57
+ Install the Hub client if needed:
58
 
59
+ ```bash
60
+ pip install huggingface_hub
61
  ```
 
 
 
 
 
 
 
 
 
 
 
 
 
62
 
63
+ Download the full repository (or only `archives/`):
 
 
 
 
64
 
65
+ ```bash
66
+ hf download hljiang/DA360 --repo-type dataset --local-dir /PATH/TO/DOWNLOAD --local-dir-use-symlinks False
67
  ```
 
 
 
 
 
 
 
 
 
 
 
68
 
69
+ Extract each benchmark into a shared `data/` folder (layout expected by [antigravity-tech/DA360](https://github.com/antigravity-tech/DA360)):
70
 
71
  ```bash
72
+ cd /PATH/TO/DOWNLOAD/archives
73
+ for z in Metropolis Matterport3D Stanford2D3D; do
74
+ unzip -q "${z}.zip" -d ../data
75
+ done
76
  ```
77
 
78
+ One-shot (download zips from the Hub and extract to `./data`):
79
 
80
  ```bash
81
+ bash -c "$(curl -fsSL https://huggingface.co/datasets/hljiang/DA360/raw/main/scripts/prepare_data.sh)"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
82
  ```
83
 
84
+ ---
85
 
86
+ ## Layout after unzip
 
 
 
87
 
88
+ ```
89
+ data/
90
+ ├── Metropolis/rgb/*.jpg + Metropolis/depth/*.png
91
+ ├── Matterport3D/{scene}/pano_skybox_color|pano_depth/
92
+ └── Stanford2D3D/area_5{a,b}/rgb_|depth/
 
 
93
  ```
94
 
95
  ---
96
 
97
+ ## Reading depth maps
98
+
99
+ | Dataset | Format | Metric depth |
100
+ | ------- | ------ | ------------- |
101
+ | Metropolis | 16-bit PNG | `depth_m = pixel_value / 512` |
102
+ | Matterport3D | 16-bit PNG | `depth_m = pixel_value / 4000` |
103
+ | Stanford2D3D | 16-bit PNG | `depth_m = pixel_value / 512` |
104
 
105
+ Reference loaders: [`demo/datasets/`](https://github.com/antigravity-tech/DA360/tree/main/demo/datasets) in the DA360 repository.
 
 
 
 
106
 
107
  ---
108
 
109
+ ## Reference
110
 
111
+ **DA360 (this release).** Jiang, H., Song, Z., Lou, Z., Xu, R., Tan, M. *Depth Anything in 360°: Towards Scale Invariance in the Wild.* arXiv:2512.22819, 2025.
 
 
 
 
 
112
 
113
+ [1] Chang, A., Dai, A., Funkhouser, T., et al. Matterport3D: Learning from RGB-D Data in Indoor Environments. *3DV*, 2017.
114
 
115
+ [2] Armeni, I., Sax, S., Zamir, A. R., Savarese, S. Joint 2D-3D-Semantic Data for Indoor Scene Understanding. *arXiv:1702.01105*, 2017.
116
 
117
+ **Metropolis raw data.** Mapillary Metropolis dataset (2024). Used under the curation pipeline described in DA360 supplementary material (Sec. S1).
118
 
119
  ---
120
 
121
+ ## Citation
122
 
123
+ If you use these benchmarks, please cite the DA360 paper and the original dataset sources above.