FaisaI commited on
Commit
c886077
·
verified ·
1 Parent(s): 0e52f81

Full build v1: 6,584 items, QA passed

Browse files
LICENSES.md ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Source licenses (verified 2026-10-29)
2
+
3
+ ## Sources used
4
+
5
+ | Source | Where | License | Redistributable | Tasks | Notes |
6
+ |---|---|---|---|---|---|
7
+ | COCO 2017 val | images.cocodataset.org | Annotations CC BY 4.0; images carry individual Flickr CC licenses | Yes, with per-image filter | cls, det, seg | Keep only images whose COCO license id is 4 (CC BY 2.0), 5 (CC BY-SA 2.0), 7 (no known copyright) or 8 (US Gov). Record the per-image license in `license`. |
8
+ | LVIS v1 val | dl.fbaipublicfiles.com/LVIS | Annotations CC BY 4.0; same COCO images | Yes, same filter | cls, det, seg | 1,203 classes, good for `class_swap_similar`. |
9
+ | Open Images V7 validation | storage.googleapis.com/openimages | Annotations CC BY 4.0; images CC BY 2.0 | Yes | cls, det, seg | Masks come from the V5 mask release (validation-masks-*.zip + CSV). Google does not guarantee each image's license; attribute per image. |
10
+ | RF100-VL | HF `probicheaux/rf100-vl` | Apache 2.0 (stated in roboflow/rf100-vl repo) | Yes | cls, det | 100 subsets across domains. Boxes only: checked 2026-10-30, the `*-seg` subsets carry no polygons. |
11
+ | PlantDoc | HF `agyaatcoder/PlantDoc` | CC BY 4.0 | Yes | cls, det | Original: github.com/pratikkayal/PlantDoc-Object-Detection-Dataset. |
12
+ | ISIC 2016 Part 1 | HF `MedOtter/ISIC2016` | CC0 1.0 | Yes | cls, det, seg | 1,279 dermoscopy images with expert lesion masks. Single class, so classification distractors come from cross-domain swaps. |
13
+ | DocLayNet v1.1 | HF `ds4sd/DocLayNet-v1.1` | CDLA-Permissive 1.0 | Yes | cls, det | Some HF mirrors tag `license: other`; the card body and the IBM page say CDLA-P 1.0. |
14
+ | PubLayNet | HF `creative-graphic-design/PubLayNet` | CDLA-Permissive 1.0 | Yes | cls, det | |
15
+
16
+ No source above is non-commercial. The dataset as a whole can be released under CC BY 4.0 with per-row attribution.
17
+
18
+ ## Dropped candidates
19
+
20
+ | Candidate | Reason |
21
+ |---|---|
22
+ | Mapillary Vistas | CC BY-NC-SA 4.0 research license |
23
+ | DOTA / iSAID | Academic use only; redistribution prohibited without permission; Google Earth terms apply |
24
+ | Kvasir-SEG | Official Simula terms: research and educational use only, commercial use forbidden |
25
+ | MVTec AD | CC BY-NC-SA 4.0 |
26
+ | NEU-DET | No license published by the original authors; mirrors disagree (MIT vs CC BY 4.0 vs "other") |
27
+ | MinneApple | CC BY-NC-SA 3.0 |
28
+ | RICO | Original terms not published; derivative CC BY 4.0 labels cover annotations only |
29
+ | BDD100K | Commercial rights reserved to BDD/BAIR members per UC Regents notice |
30
+ | Cityscapes | Non-commercial research license, registration wall |
31
+
32
+ ## Consequences for the benchmark
33
+
34
+ - Segmentation is built for 4 domains: everyday, street, medical, wildlife/underwater. Aerial, industrial and agriculture have no permissively licensed instance-mask source, and documents/UI is detection-only by spec. Record this in the dataset card.
35
+ - Optional aerial segmentation for v2: SpaceNet 2 building footprints (CC BY-SA 4.0, HF `aialliance/spacenet2`). ShareAlike would apply to that portion, so it is left out of v1.
36
+
37
+ ## Citations to include in the dataset card
38
+
39
+ lin2014coco, gupta2019lvis, kuznetsova2020openimages, robicheaux2025rf100vl, singh2020plantdoc, gutman2016isic, pfitzmann2022doclaynet, zhong2019publaynet.
README.md CHANGED
@@ -1,6 +1,10 @@
1
  ---
2
  license: cc-by-4.0
3
- pretty_name: VisVerify-Bench (preview)
 
 
 
 
4
  task_categories:
5
  - image-classification
6
  - object-detection
@@ -17,26 +21,199 @@ configs:
17
  data_files:
18
  - split: dev
19
  path: data/classification/dev-*
 
 
20
  - config_name: detection
21
  data_files:
22
  - split: dev
23
  path: data/detection/dev-*
 
 
24
  - config_name: segmentation
25
  data_files:
26
  - split: dev
27
  path: data/segmentation/dev-*
 
 
28
  ---
29
 
30
- # VisVerify-Bench (preview build)
31
 
32
- **Preview only.** This is the dev split of one domain (everyday scenes, COCO + LVIS),
33
- pushed to check the schema and the dataset viewer. The full card, count tables,
34
- license table and citations come with the full build.
35
 
36
- Each row shows a judge an image plus a *proposed* annotation and stores the
37
- ground truth. `is_correct=true` rows are untouched ground truth (TP). `is_correct=false`
38
- rows were corrupted in a controlled way (`error_type`, `severity`, `iou_with_gt`).
 
 
 
39
 
40
- - `judge_image`: the default rendering shown to a judge (zoomed crop with 50% padding for detection and segmentation; padded crop for classification)
41
- - `image`: the original image (the crop itself for classification)
42
- - `expected`: ground-truth answers for the questions in `questions.json`
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: cc-by-4.0
3
+ pretty_name: VisVerify-Bench
4
+ language:
5
+ - en
6
+ size_categories:
7
+ - 1K<n<10K
8
  task_categories:
9
  - image-classification
10
  - object-detection
 
21
  data_files:
22
  - split: dev
23
  path: data/classification/dev-*
24
+ - split: test
25
+ path: data/classification/test-*
26
  - config_name: detection
27
  data_files:
28
  - split: dev
29
  path: data/detection/dev-*
30
+ - split: test
31
+ path: data/detection/test-*
32
  - config_name: segmentation
33
  data_files:
34
  - split: dev
35
  path: data/segmentation/dev-*
36
+ - split: test
37
+ path: data/segmentation/test-*
38
  ---
39
 
40
+ # VisVerify-Bench
41
 
42
+ **Can a judge model tell a correct vision annotation from a wrong one?**
 
 
43
 
44
+ Each item shows a judge an image with a *proposed* class, box or mask. Half the items are
45
+ untouched ground truth and should be **accepted**. The other half were corrupted in a
46
+ controlled way and should be **rejected**. Every wrong item records *what* is wrong
47
+ (`error_type`) and *how badly* (`iou_with_gt`), so accuracy can be plotted against severity.
48
+ Built to compare fast "System One" decision models (JEV, Liquid d1-omni) with VLM-as-judge
49
+ baselines.
50
 
51
+ - 3 tasks (classification, detection, segmentation) x 8 domains (everyday, street, aerial, medical, industrial, agriculture, documents/UI, wildlife/underwater)
52
+ - 6,584 items: 5,988 test, 596 dev (separate photos, for prompt tuning only)
53
+ - 50% correct / 50% corrupted in every task x domain cell; 16 error types, easy / medium / hard
54
+ - Seed 42, rebuild gives an identical dataset; all sources redistributable
55
+
56
+ ## Quick start
57
+
58
+ ```python
59
+ from datasets import load_dataset
60
+ import json
61
+
62
+ ds = load_dataset("FaisaI/VisVerify-Bench", "detection", split="test")
63
+ q = json.load(open("questions.json"))["detection"] # question templates, from this repo
64
+
65
+ row = ds[0]
66
+ row["judge_image"] # crop around the proposed box, 50% padding, box + class tag drawn
67
+ row["proposed_class"] # "bus"
68
+ row["expected"] # {"class_correct": True, "box_quality": 2, "annotation_correct": False}
69
+ ```
70
+
71
+ Score answers against `expected`. Main metric: precision / recall / F1 on `annotation_correct`
72
+ with **reject as the positive class**, plus the false-alarm rate on correct items.
73
+
74
+ ## Error types
75
+
76
+ | Task | `error_type` | Corruption | Severity |
77
+ |---|---|---|---|
78
+ | all | `class_swap_random` / `_similar` / `_cross_domain` | class replaced by a random / semantically close / other-domain class | easy / hard / easy |
79
+ | det | `box_slightly_off` / `box_off` / `box_very_off` | box moved to IoU 0.75-0.85 / 0.50-0.75 / 0.20-0.50 | hard / medium / easy |
80
+ | det | `box_wrong_object` / `box_background` | box on another object / on empty background | medium / easy |
81
+ | seg | `mask_eroded` / `mask_dilated` | morphology to IoU 0.60-0.85 | medium / hard |
82
+ | seg | `mask_missing_part` / `mask_leak` / `mask_shifted` | 20-40% chunk removed / grown into background / translated to IoU 0.40-0.70 | medium / medium / easy |
83
+ | seg | `mask_wrong_instance` | mask of another instance of the same class | hard |
84
+ | det, seg | `class_and_box` / `class_and_mask` | class swap **and** a geometry error | easy |
85
+
86
+ 20% of correct boxes and masks carry a tiny jitter (IoU >= 0.92, tag `tp_jittered`) so
87
+ "pixel-perfect" cannot be used as a shortcut. Quality scores map from IoU:
88
+ 5 (>= 0.90), 4 (0.75), 3 (0.60), 2 (0.40), 1 (below).
89
+
90
+ ## Fields
91
+
92
+ `id`, `task`, `split`, `domain`, `source_dataset`, `source_id`, `license` ·
93
+ `image` (original; the crop for classification), `judge_image` (default rendering) ·
94
+ `proposed_class`, `class_description`, `gt_class` ·
95
+ `proposed_bbox`, `gt_bbox` as `[x, y, w, h]` pixels · `proposed_mask`, `gt_mask` as COCO RLE ·
96
+ `is_correct`, `error_type`, `severity`, `iou_with_gt` ·
97
+ `choice_options`, `choice_answer` (4-way question, classification) · `expected` · `tags` · `notes`
98
+
99
+ Questions live once in `questions.json`. Images are at most 1024 px on the long side; raw
100
+ geometry is kept so you can re-render (`overlay_full`, `crop_tight`, `crop_context`,
101
+ `side_by_side` in the code repo). At most 3 items per photo, no class above 10% of a domain.
102
+
103
+ ## Counts
104
+
105
+ ### classification
106
+
107
+ | domain | none | class_swap_cross_domain | class_swap_random | class_swap_similar | total |
108
+ |---|---|---|---|---|---|
109
+ | aerial | 165 | 81 | 43 | 41 | 330 |
110
+ | agriculture | 165 | 79 | 43 | 43 | 330 |
111
+ | documents_ui | 165 | 55 | 55 | 55 | 330 |
112
+ | everyday | 165 | 55 | 55 | 55 | 330 |
113
+ | industrial | 165 | 55 | 55 | 55 | 330 |
114
+ | medical | 165 | 55 | 55 | 55 | 330 |
115
+ | street | 165 | 55 | 55 | 55 | 330 |
116
+ | wildlife_underwater | 165 | 55 | 55 | 55 | 330 |
117
+ | **total** | 1320 | 490 | 416 | 414 | 2640 |
118
+
119
+ | domain | dev | test |
120
+ |---|---|---|
121
+ | aerial | 30 | 300 |
122
+ | agriculture | 30 | 300 |
123
+ | documents_ui | 30 | 300 |
124
+ | everyday | 30 | 300 |
125
+ | industrial | 30 | 300 |
126
+ | medical | 30 | 300 |
127
+ | street | 30 | 300 |
128
+ | wildlife_underwater | 30 | 300 |
129
+
130
+ Severity: easy=906, hard=414, none=1320
131
+
132
+ ### detection
133
+
134
+ | domain | none | box_background | box_off | box_slightly_off | box_very_off | box_wrong_object | class_and_box | class_swap_cross_domain | class_swap_random | class_swap_similar | total |
135
+ |---|---|---|---|---|---|---|---|---|---|---|---|
136
+ | aerial | 165 | 18 | 18 | 18 | 18 | 18 | 24 | 17 | 17 | 17 | 330 |
137
+ | agriculture | 165 | 19 | 18 | 18 | 18 | 18 | 24 | 17 | 17 | 16 | 330 |
138
+ | documents_ui | 165 | 18 | 18 | 18 | 18 | 18 | 24 | 17 | 17 | 17 | 330 |
139
+ | everyday | 165 | 18 | 18 | 18 | 18 | 18 | 24 | 17 | 17 | 17 | 330 |
140
+ | industrial | 165 | 19 | 18 | 18 | 18 | 17 | 24 | 18 | 17 | 16 | 330 |
141
+ | medical | 165 | 18 | 18 | 18 | 18 | 18 | 24 | 17 | 17 | 17 | 330 |
142
+ | street | 165 | 18 | 18 | 18 | 18 | 18 | 24 | 17 | 17 | 17 | 330 |
143
+ | wildlife_underwater | 165 | 18 | 19 | 18 | 18 | 17 | 24 | 18 | 17 | 16 | 330 |
144
+ | **total** | 1320 | 146 | 145 | 144 | 144 | 142 | 192 | 138 | 136 | 133 | 2640 |
145
+
146
+ | domain | dev | test |
147
+ |---|---|---|
148
+ | aerial | 30 | 300 |
149
+ | agriculture | 30 | 300 |
150
+ | documents_ui | 30 | 300 |
151
+ | everyday | 30 | 300 |
152
+ | industrial | 30 | 300 |
153
+ | medical | 30 | 300 |
154
+ | street | 30 | 300 |
155
+ | wildlife_underwater | 30 | 300 |
156
+
157
+ Severity: easy=756, hard=277, medium=287, none=1320
158
+
159
+ ### segmentation
160
+
161
+ | domain | none | class_and_mask | class_swap_cross_domain | class_swap_random | class_swap_similar | mask_dilated | mask_eroded | mask_leak | mask_missing_part | mask_shifted | mask_wrong_instance | total |
162
+ |---|---|---|---|---|---|---|---|---|---|---|---|---|
163
+ | everyday | 165 | 24 | 17 | 17 | 16 | 16 | 15 | 15 | 15 | 15 | 15 | 330 |
164
+ | medical | 165 | 0 | 28 | 0 | 0 | 28 | 28 | 27 | 27 | 27 | 0 | 330 |
165
+ | street | 157 | 23 | 16 | 16 | 16 | 15 | 15 | 15 | 14 | 14 | 13 | 314 |
166
+ | wildlife_underwater | 165 | 24 | 17 | 17 | 16 | 16 | 15 | 15 | 15 | 15 | 15 | 330 |
167
+ | **total** | 652 | 71 | 78 | 50 | 48 | 75 | 73 | 72 | 71 | 71 | 43 | 1304 |
168
+
169
+ | domain | dev | test |
170
+ |---|---|---|
171
+ | everyday | 30 | 300 |
172
+ | medical | 30 | 300 |
173
+ | street | 26 | 288 |
174
+ | wildlife_underwater | 30 | 300 |
175
+
176
+ Severity: easy=270, hard=166, medium=216, none=652
177
+
178
+ ## Sources and licenses
179
+
180
+ | Source | Where | License | Redistributable | Tasks | Notes |
181
+ |---|---|---|---|---|---|
182
+ | COCO 2017 val | images.cocodataset.org | Annotations CC BY 4.0; images carry individual Flickr CC licenses | Yes, with per-image filter | cls, det, seg | Keep only images whose COCO license id is 4 (CC BY 2.0), 5 (CC BY-SA 2.0), 7 (no known copyright) or 8 (US Gov). Record the per-image license in `license`. |
183
+ | LVIS v1 val | dl.fbaipublicfiles.com/LVIS | Annotations CC BY 4.0; same COCO images | Yes, same filter | cls, det, seg | 1,203 classes, good for `class_swap_similar`. |
184
+ | Open Images V7 validation | storage.googleapis.com/openimages | Annotations CC BY 4.0; images CC BY 2.0 | Yes | cls, det, seg | Masks come from the V5 mask release (validation-masks-*.zip + CSV). Google does not guarantee each image's license; attribute per image. |
185
+ | RF100-VL | HF `probicheaux/rf100-vl` | Apache 2.0 (stated in roboflow/rf100-vl repo) | Yes | cls, det | 100 subsets across domains. Boxes only: checked 2026-10-30, the `*-seg` subsets carry no polygons. |
186
+ | PlantDoc | HF `agyaatcoder/PlantDoc` | CC BY 4.0 | Yes | cls, det | Original: github.com/pratikkayal/PlantDoc-Object-Detection-Dataset. |
187
+ | ISIC 2016 Part 1 | HF `MedOtter/ISIC2016` | CC0 1.0 | Yes | cls, det, seg | 1,279 dermoscopy images with expert lesion masks. Single class, so classification distractors come from cross-domain swaps. |
188
+ | DocLayNet v1.1 | HF `ds4sd/DocLayNet-v1.1` | CDLA-Permissive 1.0 | Yes | cls, det | Some HF mirrors tag `license: other`; the card body and the IBM page say CDLA-P 1.0. |
189
+ | PubLayNet | HF `creative-graphic-design/PubLayNet` | CDLA-Permissive 1.0 | Yes | cls, det | |
190
+
191
+ No non-commercial source is used. COCO and LVIS photos keep their individual Flickr licenses
192
+ (only CC BY, CC BY-SA, public domain and US Government photos were used) and each row
193
+ records its own. Segmentation covers 4 domains; aerial, industrial and agriculture had no
194
+ permissively licensed mask source, and documents is detection-only.
195
+
196
+ ## Limitations
197
+
198
+ - Cells below target: everyday/classification (330), everyday/detection (330), everyday/segmentation (330), street/classification (330), street/detection (330), street/segmentation (314), aerial/classification (330), aerial/detection (330), medical/classification (330), medical/detection (330), medical/segmentation (330), industrial/classification (330), industrial/detection (330), agriculture/classification (330), agriculture/detection (330), documents_ui/classification (330), documents_ui/detection (330), wildlife_underwater/classification (330), wildlife_underwater/detection (330), wildlife_underwater/segmentation (330).
199
+
200
+ - Small: 150 items per task and domain. Good for ranking judges, not for per-class claims.
201
+ - Synthetic corruptions are cleaner than real annotator mistakes.
202
+ - Correct items inherit any label noise in the source datasets.
203
+ - ISIC 2016 is single-class, so its classification items use only random and cross-domain swaps.
204
+ - Street has no driving-camera dataset (Mapillary, BDD100K, Cityscapes failed the license check).
205
+
206
+ ## Citation
207
+
208
+ ```bibtex
209
+ @misc{visverifybench2026,
210
+ title = {VisVerify-Bench: Can Fast Judges Verify Vision Annotations?},
211
+ author = {Faisal},
212
+ year = {2026},
213
+ url = {https://huggingface.co/datasets/FaisaI/VisVerify-Bench}
214
+ }
215
+ ```
216
+
217
+ Sources: COCO (Lin et al. 2014), LVIS (Gupta et al. 2019), Open Images (Kuznetsova et al.
218
+ 2020), Roboflow100-VL (Robicheaux et al. 2025), PlantDoc (Singh et al. 2020), ISIC 2016
219
+ (Gutman et al. 2016), DocLayNet (Pfitzmann et al. 2022), PubLayNet (Zhong et al. 2019).
build_report.json CHANGED
@@ -1,45 +1,449 @@
1
  {
2
  "cells": {
3
  "everyday/classification": {
4
- "n": 15,
5
  "by_split_error": {
6
- "dev/class_swap_random": 4,
7
- "dev/class_swap_similar": 3,
8
- "dev/none": 8
 
 
 
 
 
9
  }
10
  },
11
  "everyday/detection": {
12
- "n": 15,
13
  "by_split_error": {
14
- "dev/box_background": 1,
15
- "dev/box_off": 1,
16
- "dev/box_slightly_off": 1,
17
- "dev/box_very_off": 1,
18
- "dev/box_wrong_object": 1,
19
- "dev/class_and_box": 1,
 
20
  "dev/class_swap_random": 1,
21
- "dev/none": 8
 
 
 
 
 
 
 
 
 
 
 
22
  }
23
  },
24
  "everyday/segmentation": {
25
- "n": 15,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
26
  "by_split_error": {
27
- "dev/class_and_mask": 1,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
28
  "dev/class_swap_random": 1,
29
  "dev/class_swap_similar": 1,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
30
  "dev/mask_dilated": 1,
31
  "dev/mask_eroded": 1,
32
  "dev/mask_leak": 1,
33
  "dev/mask_missing_part": 1,
34
- "dev/none": 8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
35
  }
36
  }
37
  },
38
- "skipped": {},
 
 
 
 
 
 
 
 
 
 
 
39
  "timings": {
40
- "load+fit/everyday": 218.1,
41
- "build/everyday/classification": 0.0,
42
- "build/everyday/detection": 0.1,
43
- "build/everyday/segmentation": 0.1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44
  }
45
  }
 
1
  {
2
  "cells": {
3
  "everyday/classification": {
4
+ "n": 330,
5
  "by_split_error": {
6
+ "dev/class_swap_cross_domain": 5,
7
+ "dev/class_swap_random": 5,
8
+ "dev/class_swap_similar": 5,
9
+ "dev/none": 15,
10
+ "test/class_swap_cross_domain": 50,
11
+ "test/class_swap_random": 50,
12
+ "test/class_swap_similar": 50,
13
+ "test/none": 150
14
  }
15
  },
16
  "everyday/detection": {
17
+ "n": 330,
18
  "by_split_error": {
19
+ "dev/box_background": 2,
20
+ "dev/box_off": 2,
21
+ "dev/box_slightly_off": 2,
22
+ "dev/box_very_off": 2,
23
+ "dev/box_wrong_object": 2,
24
+ "dev/class_and_box": 2,
25
+ "dev/class_swap_cross_domain": 1,
26
  "dev/class_swap_random": 1,
27
+ "dev/class_swap_similar": 1,
28
+ "dev/none": 15,
29
+ "test/box_background": 16,
30
+ "test/box_off": 16,
31
+ "test/box_slightly_off": 16,
32
+ "test/box_very_off": 16,
33
+ "test/box_wrong_object": 16,
34
+ "test/class_and_box": 22,
35
+ "test/class_swap_cross_domain": 16,
36
+ "test/class_swap_random": 16,
37
+ "test/class_swap_similar": 16,
38
+ "test/none": 150
39
  }
40
  },
41
  "everyday/segmentation": {
42
+ "n": 330,
43
+ "by_split_error": {
44
+ "dev/class_and_mask": 2,
45
+ "dev/class_swap_cross_domain": 2,
46
+ "dev/class_swap_random": 2,
47
+ "dev/class_swap_similar": 2,
48
+ "dev/mask_dilated": 2,
49
+ "dev/mask_eroded": 1,
50
+ "dev/mask_leak": 1,
51
+ "dev/mask_missing_part": 1,
52
+ "dev/mask_shifted": 1,
53
+ "dev/mask_wrong_instance": 1,
54
+ "dev/none": 15,
55
+ "test/class_and_mask": 22,
56
+ "test/class_swap_cross_domain": 15,
57
+ "test/class_swap_random": 15,
58
+ "test/class_swap_similar": 14,
59
+ "test/mask_dilated": 14,
60
+ "test/mask_eroded": 14,
61
+ "test/mask_leak": 14,
62
+ "test/mask_missing_part": 14,
63
+ "test/mask_shifted": 14,
64
+ "test/mask_wrong_instance": 14,
65
+ "test/none": 150
66
+ }
67
+ },
68
+ "street/classification": {
69
+ "n": 330,
70
  "by_split_error": {
71
+ "dev/class_swap_cross_domain": 5,
72
+ "dev/class_swap_random": 5,
73
+ "dev/class_swap_similar": 5,
74
+ "dev/none": 15,
75
+ "test/class_swap_cross_domain": 50,
76
+ "test/class_swap_random": 50,
77
+ "test/class_swap_similar": 50,
78
+ "test/none": 150
79
+ }
80
+ },
81
+ "street/detection": {
82
+ "n": 330,
83
+ "by_split_error": {
84
+ "dev/box_background": 2,
85
+ "dev/box_off": 2,
86
+ "dev/box_slightly_off": 2,
87
+ "dev/box_very_off": 2,
88
+ "dev/box_wrong_object": 2,
89
+ "dev/class_and_box": 2,
90
+ "dev/class_swap_cross_domain": 1,
91
  "dev/class_swap_random": 1,
92
  "dev/class_swap_similar": 1,
93
+ "dev/none": 15,
94
+ "test/box_background": 16,
95
+ "test/box_off": 16,
96
+ "test/box_slightly_off": 16,
97
+ "test/box_very_off": 16,
98
+ "test/box_wrong_object": 16,
99
+ "test/class_and_box": 22,
100
+ "test/class_swap_cross_domain": 16,
101
+ "test/class_swap_random": 16,
102
+ "test/class_swap_similar": 16,
103
+ "test/none": 150
104
+ }
105
+ },
106
+ "street/segmentation": {
107
+ "n": 314,
108
+ "by_split_error": {
109
+ "dev/class_and_mask": 2,
110
+ "dev/class_swap_cross_domain": 2,
111
+ "dev/class_swap_random": 2,
112
+ "dev/class_swap_similar": 2,
113
  "dev/mask_dilated": 1,
114
  "dev/mask_eroded": 1,
115
  "dev/mask_leak": 1,
116
  "dev/mask_missing_part": 1,
117
+ "dev/mask_shifted": 1,
118
+ "dev/none": 13,
119
+ "test/class_and_mask": 21,
120
+ "test/class_swap_cross_domain": 14,
121
+ "test/class_swap_random": 14,
122
+ "test/class_swap_similar": 14,
123
+ "test/mask_dilated": 14,
124
+ "test/mask_eroded": 14,
125
+ "test/mask_leak": 14,
126
+ "test/mask_missing_part": 13,
127
+ "test/mask_shifted": 13,
128
+ "test/mask_wrong_instance": 13,
129
+ "test/none": 144
130
+ }
131
+ },
132
+ "aerial/classification": {
133
+ "n": 330,
134
+ "by_split_error": {
135
+ "dev/class_swap_cross_domain": 8,
136
+ "dev/class_swap_random": 4,
137
+ "dev/class_swap_similar": 3,
138
+ "dev/none": 15,
139
+ "test/class_swap_cross_domain": 73,
140
+ "test/class_swap_random": 39,
141
+ "test/class_swap_similar": 38,
142
+ "test/none": 150
143
+ }
144
+ },
145
+ "aerial/detection": {
146
+ "n": 330,
147
+ "by_split_error": {
148
+ "dev/box_background": 2,
149
+ "dev/box_off": 2,
150
+ "dev/box_slightly_off": 2,
151
+ "dev/box_very_off": 2,
152
+ "dev/box_wrong_object": 2,
153
+ "dev/class_and_box": 2,
154
+ "dev/class_swap_cross_domain": 1,
155
+ "dev/class_swap_random": 1,
156
+ "dev/class_swap_similar": 1,
157
+ "dev/none": 15,
158
+ "test/box_background": 16,
159
+ "test/box_off": 16,
160
+ "test/box_slightly_off": 16,
161
+ "test/box_very_off": 16,
162
+ "test/box_wrong_object": 16,
163
+ "test/class_and_box": 22,
164
+ "test/class_swap_cross_domain": 16,
165
+ "test/class_swap_random": 16,
166
+ "test/class_swap_similar": 16,
167
+ "test/none": 150
168
+ }
169
+ },
170
+ "medical/classification": {
171
+ "n": 330,
172
+ "by_split_error": {
173
+ "dev/class_swap_cross_domain": 5,
174
+ "dev/class_swap_random": 5,
175
+ "dev/class_swap_similar": 5,
176
+ "dev/none": 15,
177
+ "test/class_swap_cross_domain": 50,
178
+ "test/class_swap_random": 50,
179
+ "test/class_swap_similar": 50,
180
+ "test/none": 150
181
+ }
182
+ },
183
+ "medical/detection": {
184
+ "n": 330,
185
+ "by_split_error": {
186
+ "dev/box_background": 2,
187
+ "dev/box_off": 2,
188
+ "dev/box_slightly_off": 2,
189
+ "dev/box_very_off": 2,
190
+ "dev/box_wrong_object": 2,
191
+ "dev/class_and_box": 2,
192
+ "dev/class_swap_cross_domain": 1,
193
+ "dev/class_swap_random": 1,
194
+ "dev/class_swap_similar": 1,
195
+ "dev/none": 15,
196
+ "test/box_background": 16,
197
+ "test/box_off": 16,
198
+ "test/box_slightly_off": 16,
199
+ "test/box_very_off": 16,
200
+ "test/box_wrong_object": 16,
201
+ "test/class_and_box": 22,
202
+ "test/class_swap_cross_domain": 16,
203
+ "test/class_swap_random": 16,
204
+ "test/class_swap_similar": 16,
205
+ "test/none": 150
206
+ }
207
+ },
208
+ "medical/segmentation": {
209
+ "n": 330,
210
+ "by_split_error": {
211
+ "dev/class_swap_cross_domain": 3,
212
+ "dev/mask_dilated": 3,
213
+ "dev/mask_eroded": 3,
214
+ "dev/mask_leak": 2,
215
+ "dev/mask_missing_part": 2,
216
+ "dev/mask_shifted": 2,
217
+ "dev/none": 15,
218
+ "test/class_swap_cross_domain": 25,
219
+ "test/mask_dilated": 25,
220
+ "test/mask_eroded": 25,
221
+ "test/mask_leak": 25,
222
+ "test/mask_missing_part": 25,
223
+ "test/mask_shifted": 25,
224
+ "test/none": 150
225
+ }
226
+ },
227
+ "industrial/classification": {
228
+ "n": 330,
229
+ "by_split_error": {
230
+ "dev/class_swap_cross_domain": 5,
231
+ "dev/class_swap_random": 5,
232
+ "dev/class_swap_similar": 5,
233
+ "dev/none": 15,
234
+ "test/class_swap_cross_domain": 50,
235
+ "test/class_swap_random": 50,
236
+ "test/class_swap_similar": 50,
237
+ "test/none": 150
238
+ }
239
+ },
240
+ "industrial/detection": {
241
+ "n": 330,
242
+ "by_split_error": {
243
+ "dev/box_background": 2,
244
+ "dev/box_off": 2,
245
+ "dev/box_slightly_off": 2,
246
+ "dev/box_very_off": 2,
247
+ "dev/box_wrong_object": 1,
248
+ "dev/class_and_box": 2,
249
+ "dev/class_swap_cross_domain": 2,
250
+ "dev/class_swap_random": 1,
251
+ "dev/class_swap_similar": 1,
252
+ "dev/none": 15,
253
+ "test/box_background": 17,
254
+ "test/box_off": 16,
255
+ "test/box_slightly_off": 16,
256
+ "test/box_very_off": 16,
257
+ "test/box_wrong_object": 16,
258
+ "test/class_and_box": 22,
259
+ "test/class_swap_cross_domain": 16,
260
+ "test/class_swap_random": 16,
261
+ "test/class_swap_similar": 15,
262
+ "test/none": 150
263
+ }
264
+ },
265
+ "agriculture/classification": {
266
+ "n": 330,
267
+ "by_split_error": {
268
+ "dev/class_swap_cross_domain": 11,
269
+ "dev/class_swap_random": 2,
270
+ "dev/class_swap_similar": 2,
271
+ "dev/none": 15,
272
+ "test/class_swap_cross_domain": 68,
273
+ "test/class_swap_random": 41,
274
+ "test/class_swap_similar": 41,
275
+ "test/none": 150
276
+ }
277
+ },
278
+ "agriculture/detection": {
279
+ "n": 330,
280
+ "by_split_error": {
281
+ "dev/box_background": 2,
282
+ "dev/box_off": 2,
283
+ "dev/box_slightly_off": 2,
284
+ "dev/box_very_off": 2,
285
+ "dev/box_wrong_object": 2,
286
+ "dev/class_and_box": 2,
287
+ "dev/class_swap_cross_domain": 1,
288
+ "dev/class_swap_random": 1,
289
+ "dev/class_swap_similar": 1,
290
+ "dev/none": 15,
291
+ "test/box_background": 17,
292
+ "test/box_off": 16,
293
+ "test/box_slightly_off": 16,
294
+ "test/box_very_off": 16,
295
+ "test/box_wrong_object": 16,
296
+ "test/class_and_box": 22,
297
+ "test/class_swap_cross_domain": 16,
298
+ "test/class_swap_random": 16,
299
+ "test/class_swap_similar": 15,
300
+ "test/none": 150
301
+ }
302
+ },
303
+ "documents_ui/classification": {
304
+ "n": 330,
305
+ "by_split_error": {
306
+ "dev/class_swap_cross_domain": 5,
307
+ "dev/class_swap_random": 5,
308
+ "dev/class_swap_similar": 5,
309
+ "dev/none": 15,
310
+ "test/class_swap_cross_domain": 50,
311
+ "test/class_swap_random": 50,
312
+ "test/class_swap_similar": 50,
313
+ "test/none": 150
314
+ }
315
+ },
316
+ "documents_ui/detection": {
317
+ "n": 330,
318
+ "by_split_error": {
319
+ "dev/box_background": 2,
320
+ "dev/box_off": 2,
321
+ "dev/box_slightly_off": 2,
322
+ "dev/box_very_off": 2,
323
+ "dev/box_wrong_object": 2,
324
+ "dev/class_and_box": 2,
325
+ "dev/class_swap_cross_domain": 1,
326
+ "dev/class_swap_random": 1,
327
+ "dev/class_swap_similar": 1,
328
+ "dev/none": 15,
329
+ "test/box_background": 16,
330
+ "test/box_off": 16,
331
+ "test/box_slightly_off": 16,
332
+ "test/box_very_off": 16,
333
+ "test/box_wrong_object": 16,
334
+ "test/class_and_box": 22,
335
+ "test/class_swap_cross_domain": 16,
336
+ "test/class_swap_random": 16,
337
+ "test/class_swap_similar": 16,
338
+ "test/none": 150
339
+ }
340
+ },
341
+ "wildlife_underwater/classification": {
342
+ "n": 330,
343
+ "by_split_error": {
344
+ "dev/class_swap_cross_domain": 5,
345
+ "dev/class_swap_random": 5,
346
+ "dev/class_swap_similar": 5,
347
+ "dev/none": 15,
348
+ "test/class_swap_cross_domain": 50,
349
+ "test/class_swap_random": 50,
350
+ "test/class_swap_similar": 50,
351
+ "test/none": 150
352
+ }
353
+ },
354
+ "wildlife_underwater/detection": {
355
+ "n": 330,
356
+ "by_split_error": {
357
+ "dev/box_background": 2,
358
+ "dev/box_off": 2,
359
+ "dev/box_slightly_off": 2,
360
+ "dev/box_very_off": 2,
361
+ "dev/box_wrong_object": 1,
362
+ "dev/class_and_box": 2,
363
+ "dev/class_swap_cross_domain": 2,
364
+ "dev/class_swap_random": 1,
365
+ "dev/class_swap_similar": 1,
366
+ "dev/none": 15,
367
+ "test/box_background": 16,
368
+ "test/box_off": 17,
369
+ "test/box_slightly_off": 16,
370
+ "test/box_very_off": 16,
371
+ "test/box_wrong_object": 16,
372
+ "test/class_and_box": 22,
373
+ "test/class_swap_cross_domain": 16,
374
+ "test/class_swap_random": 16,
375
+ "test/class_swap_similar": 15,
376
+ "test/none": 150
377
+ }
378
+ },
379
+ "wildlife_underwater/segmentation": {
380
+ "n": 330,
381
+ "by_split_error": {
382
+ "dev/class_and_mask": 2,
383
+ "dev/class_swap_cross_domain": 2,
384
+ "dev/class_swap_random": 2,
385
+ "dev/class_swap_similar": 2,
386
+ "dev/mask_dilated": 2,
387
+ "dev/mask_eroded": 1,
388
+ "dev/mask_leak": 1,
389
+ "dev/mask_missing_part": 1,
390
+ "dev/mask_shifted": 1,
391
+ "dev/mask_wrong_instance": 1,
392
+ "dev/none": 15,
393
+ "test/class_and_mask": 22,
394
+ "test/class_swap_cross_domain": 15,
395
+ "test/class_swap_random": 15,
396
+ "test/class_swap_similar": 14,
397
+ "test/mask_dilated": 14,
398
+ "test/mask_eroded": 14,
399
+ "test/mask_leak": 14,
400
+ "test/mask_missing_part": 14,
401
+ "test/mask_shifted": 14,
402
+ "test/mask_wrong_instance": 14,
403
+ "test/none": 150
404
  }
405
  }
406
  },
407
+ "skipped": {
408
+ "everyday/near_duplicates_merged": 1208,
409
+ "street/near_duplicates_merged": 786,
410
+ "street/segmentation/dev": 4,
411
+ "street/segmentation/test": 12,
412
+ "aerial/near_duplicates_merged": 301,
413
+ "medical/near_duplicates_merged": 61,
414
+ "industrial/near_duplicates_merged": 411,
415
+ "agriculture/near_duplicates_merged": 1197,
416
+ "documents_ui/near_duplicates_merged": 319,
417
+ "wildlife_underwater/near_duplicates_merged": 528
418
+ },
419
  "timings": {
420
+ "load+fit/everyday": 0.1,
421
+ "load+fit/street": 251.7,
422
+ "load+fit/aerial": 0.1,
423
+ "load+fit/medical": 0.1,
424
+ "load+fit/industrial": 0.1,
425
+ "load+fit/agriculture": 0.3,
426
+ "load+fit/documents_ui": 0.2,
427
+ "load+fit/wildlife_underwater": 342.5,
428
+ "build/everyday/classification": 6.9,
429
+ "build/everyday/detection": 1.1,
430
+ "build/everyday/segmentation": 3.2,
431
+ "build/street/classification": 14.3,
432
+ "build/street/detection": 1.3,
433
+ "build/street/segmentation": 214.0,
434
+ "build/aerial/classification": 6.3,
435
+ "build/aerial/detection": 1.5,
436
+ "build/medical/classification": 5.6,
437
+ "build/medical/detection": 1.1,
438
+ "build/medical/segmentation": 616.2,
439
+ "build/industrial/classification": 6.0,
440
+ "build/industrial/detection": 1.3,
441
+ "build/agriculture/classification": 13.4,
442
+ "build/agriculture/detection": 1.5,
443
+ "build/documents_ui/classification": 15.1,
444
+ "build/documents_ui/detection": 1.3,
445
+ "build/wildlife_underwater/classification": 17.8,
446
+ "build/wildlife_underwater/detection": 1.4,
447
+ "build/wildlife_underwater/segmentation": 39.5
448
  }
449
  }
data/classification/dev-00000-of-00001.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7175e03bdf4a36023f9ab1b6f696e53b65862cd05cf690e10137953e9f102cab
3
- size 851145
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:742720627c883c57d7b9f4ef9caced22ebd4a8cea25940fdb3e3f9da5b960cab
3
+ size 11129183
data/classification/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cf54daaa6c58cb7374fb1a2e02607d0e3f4fe2c6bd4e15384db4353bc14c7253
3
+ size 109779103
data/detection/dev-00000-of-00001.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6ec6be4988b821fe9cc7c0127923538b7984320554befe159071d467ab58f89a
3
- size 2486505
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:31b546472ec7fc0006b005fa9b8ead32991e1b56fe684b0ca0097e1fe2cd2d87
3
+ size 41986371
data/detection/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0e326e061619241715234c5c9ec9573eaeda70e3001351dcc813075076b7ae08
3
+ size 408865319
data/segmentation/dev-00000-of-00001.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:babd8709c386bb96bcdc1c1b640a484cc6d4df89f35304a22c5f3927835dd533
3
- size 1922013
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0b2fe0485a5e0a9627044c9a9a19ca7e250f5791a7c61c2cbeb8b791e3bfc9de
3
+ size 28205657
data/segmentation/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:598c7e357e5b7a38dd8c3657776c973976b7dfa7ca54dde7098be9a69d861a4e
3
+ size 289709189
qa_report.json CHANGED
@@ -1,29 +1,214 @@
1
  {
2
  "rows": {
3
- "classification": 15,
4
- "detection": 15,
5
- "segmentation": 15
6
  },
7
  "ratio": {
8
  "classification": {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9
  "everyday/dev": {
10
- "tp": 8,
11
- "n": 15,
12
- "tp_fraction": 0.533
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
13
  }
14
  },
15
  "detection": {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
16
  "everyday/dev": {
17
- "tp": 8,
18
- "n": 15,
19
- "tp_fraction": 0.533
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
20
  }
21
  },
22
  "segmentation": {
23
  "everyday/dev": {
24
- "tp": 8,
25
- "n": 15,
26
- "tp_fraction": 0.533
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
27
  }
28
  }
29
  },
@@ -39,13 +224,16 @@
39
  },
40
  "duplicates": {
41
  "classification": {
42
- "dev_test_collisions": 0
 
43
  },
44
  "detection": {
45
- "dev_test_collisions": 0
 
46
  },
47
  "segmentation": {
48
- "dev_test_collisions": 0
 
49
  }
50
  },
51
  "problems": [],
 
1
  {
2
  "rows": {
3
+ "classification": 2640,
4
+ "detection": 2640,
5
+ "segmentation": 1304
6
  },
7
  "ratio": {
8
  "classification": {
9
+ "aerial/dev": {
10
+ "tp": 15,
11
+ "n": 30,
12
+ "tp_fraction": 0.5
13
+ },
14
+ "aerial/test": {
15
+ "tp": 150,
16
+ "n": 300,
17
+ "tp_fraction": 0.5
18
+ },
19
+ "agriculture/dev": {
20
+ "tp": 15,
21
+ "n": 30,
22
+ "tp_fraction": 0.5
23
+ },
24
+ "agriculture/test": {
25
+ "tp": 150,
26
+ "n": 300,
27
+ "tp_fraction": 0.5
28
+ },
29
+ "documents_ui/dev": {
30
+ "tp": 15,
31
+ "n": 30,
32
+ "tp_fraction": 0.5
33
+ },
34
+ "documents_ui/test": {
35
+ "tp": 150,
36
+ "n": 300,
37
+ "tp_fraction": 0.5
38
+ },
39
  "everyday/dev": {
40
+ "tp": 15,
41
+ "n": 30,
42
+ "tp_fraction": 0.5
43
+ },
44
+ "everyday/test": {
45
+ "tp": 150,
46
+ "n": 300,
47
+ "tp_fraction": 0.5
48
+ },
49
+ "industrial/dev": {
50
+ "tp": 15,
51
+ "n": 30,
52
+ "tp_fraction": 0.5
53
+ },
54
+ "industrial/test": {
55
+ "tp": 150,
56
+ "n": 300,
57
+ "tp_fraction": 0.5
58
+ },
59
+ "medical/dev": {
60
+ "tp": 15,
61
+ "n": 30,
62
+ "tp_fraction": 0.5
63
+ },
64
+ "medical/test": {
65
+ "tp": 150,
66
+ "n": 300,
67
+ "tp_fraction": 0.5
68
+ },
69
+ "street/dev": {
70
+ "tp": 15,
71
+ "n": 30,
72
+ "tp_fraction": 0.5
73
+ },
74
+ "street/test": {
75
+ "tp": 150,
76
+ "n": 300,
77
+ "tp_fraction": 0.5
78
+ },
79
+ "wildlife_underwater/dev": {
80
+ "tp": 15,
81
+ "n": 30,
82
+ "tp_fraction": 0.5
83
+ },
84
+ "wildlife_underwater/test": {
85
+ "tp": 150,
86
+ "n": 300,
87
+ "tp_fraction": 0.5
88
  }
89
  },
90
  "detection": {
91
+ "aerial/dev": {
92
+ "tp": 15,
93
+ "n": 30,
94
+ "tp_fraction": 0.5
95
+ },
96
+ "aerial/test": {
97
+ "tp": 150,
98
+ "n": 300,
99
+ "tp_fraction": 0.5
100
+ },
101
+ "agriculture/dev": {
102
+ "tp": 15,
103
+ "n": 30,
104
+ "tp_fraction": 0.5
105
+ },
106
+ "agriculture/test": {
107
+ "tp": 150,
108
+ "n": 300,
109
+ "tp_fraction": 0.5
110
+ },
111
+ "documents_ui/dev": {
112
+ "tp": 15,
113
+ "n": 30,
114
+ "tp_fraction": 0.5
115
+ },
116
+ "documents_ui/test": {
117
+ "tp": 150,
118
+ "n": 300,
119
+ "tp_fraction": 0.5
120
+ },
121
  "everyday/dev": {
122
+ "tp": 15,
123
+ "n": 30,
124
+ "tp_fraction": 0.5
125
+ },
126
+ "everyday/test": {
127
+ "tp": 150,
128
+ "n": 300,
129
+ "tp_fraction": 0.5
130
+ },
131
+ "industrial/dev": {
132
+ "tp": 15,
133
+ "n": 30,
134
+ "tp_fraction": 0.5
135
+ },
136
+ "industrial/test": {
137
+ "tp": 150,
138
+ "n": 300,
139
+ "tp_fraction": 0.5
140
+ },
141
+ "medical/dev": {
142
+ "tp": 15,
143
+ "n": 30,
144
+ "tp_fraction": 0.5
145
+ },
146
+ "medical/test": {
147
+ "tp": 150,
148
+ "n": 300,
149
+ "tp_fraction": 0.5
150
+ },
151
+ "street/dev": {
152
+ "tp": 15,
153
+ "n": 30,
154
+ "tp_fraction": 0.5
155
+ },
156
+ "street/test": {
157
+ "tp": 150,
158
+ "n": 300,
159
+ "tp_fraction": 0.5
160
+ },
161
+ "wildlife_underwater/dev": {
162
+ "tp": 15,
163
+ "n": 30,
164
+ "tp_fraction": 0.5
165
+ },
166
+ "wildlife_underwater/test": {
167
+ "tp": 150,
168
+ "n": 300,
169
+ "tp_fraction": 0.5
170
  }
171
  },
172
  "segmentation": {
173
  "everyday/dev": {
174
+ "tp": 15,
175
+ "n": 30,
176
+ "tp_fraction": 0.5
177
+ },
178
+ "everyday/test": {
179
+ "tp": 150,
180
+ "n": 300,
181
+ "tp_fraction": 0.5
182
+ },
183
+ "medical/dev": {
184
+ "tp": 15,
185
+ "n": 30,
186
+ "tp_fraction": 0.5
187
+ },
188
+ "medical/test": {
189
+ "tp": 150,
190
+ "n": 300,
191
+ "tp_fraction": 0.5
192
+ },
193
+ "street/dev": {
194
+ "tp": 13,
195
+ "n": 26,
196
+ "tp_fraction": 0.5
197
+ },
198
+ "street/test": {
199
+ "tp": 144,
200
+ "n": 288,
201
+ "tp_fraction": 0.5
202
+ },
203
+ "wildlife_underwater/dev": {
204
+ "tp": 15,
205
+ "n": 30,
206
+ "tp_fraction": 0.5
207
+ },
208
+ "wildlife_underwater/test": {
209
+ "tp": 150,
210
+ "n": 300,
211
+ "tp_fraction": 0.5
212
  }
213
  }
214
  },
 
224
  },
225
  "duplicates": {
226
  "classification": {
227
+ "shared_photo_keys": 0,
228
+ "dev_test_phash_collisions": 0
229
  },
230
  "detection": {
231
+ "shared_photo_keys": 0,
232
+ "dev_test_phash_collisions": 0
233
  },
234
  "segmentation": {
235
+ "shared_photo_keys": 0,
236
+ "dev_test_phash_collisions": 0
237
  }
238
  },
239
  "problems": [],