COCO-Stuff 164K
Models label each pixel with an object or background category. The benchmark adds annotations for regions such as sky, grass, and pavement.
- Made by
- Holger Caesar and colleagues
- Kind
- Detection and captioning
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Items from
- Another benchmark
- Refresh
- One fixed set
- Run by
- Anyone
- Based on
- COCO
- In views
- Vision, Image, video and 3D generation
SAN ViT-L_14 tops the leaderboard with 45.8%
Best score per entry from MMSegmentation’s leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.
- 45.8 SAN ViT-L_14
- 43.4 DeepLabV3 R-101-D8
- 42.4 PSPNet R-101-D8
- 42.2 DeepLabV3 R-50-D8
- 41.9 SAN ViT-B_16
- 40.8 PSPNet R-50-D8
- 38.0 BiSeNetV1 R-101-D32
- 35.4 BiSeNetV1 R-50-D32
- 31.8 BiSeNetV1 R-101-D32 (No Pretrain)
- 30.3 BiSeNetV1 R-50-D32 (No Pretrain)
- 29.3 BiSeNetV1 R-18-D32
- 26.1 BiSeNetV1 R-18-D32 (No Pretrain)
- 1 SAN ViT-L_14 60000 iterations, 640x640 crop, CLIP_ViT-L14 pretraining, Single scale 45.8%
- 2 DeepLabV3 R-101-D8 320000 iterations, 512x512 crop, Multiscale and flip 43.4%
- 3 PSPNet R-101-D8 320000 iterations, 512x512 crop, Multiscale and flip 42.4%
- 4 DeepLabV3 R-50-D8 320000 iterations, 512x512 crop, Multiscale and flip 42.2%
- 5 SAN ViT-B_16 60000 iterations, 640x640 crop, CLIP_ViT-B16 pretraining, Single scale 41.9%
- 6 PSPNet R-50-D8 320000 iterations, 512x512 crop, Multiscale and flip 40.8%
- 7 BiSeNetV1 R-101-D32 160000 iterations, 512x512 crop, Multiscale and flip 38%
- 8 BiSeNetV1 R-50-D32 160000 iterations, 512x512 crop, Multiscale and flip 35.4%
- 9 BiSeNetV1 R-101-D32 (No Pretrain) 160000 iterations, 512x512 crop, Multiscale and flip 31.8%
- 10 BiSeNetV1 R-50-D32 (No Pretrain) 160000 iterations, 512x512 crop, Multiscale and flip 30.3%
- 11 BiSeNetV1 R-18-D32 160000 iterations, 512x512 crop, Multiscale and flip 29.3%
- 12 BiSeNetV1 R-18-D32 (No Pretrain) 160000 iterations, 512x512 crop, Multiscale and flip 26.1%
Source: MMSegmentation’s leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | SAN ViT-L_14 | 45.8% | 60000 iterations, 640x640 crop, CLIP_ViT-L14 pretraining, Single scale | Official leaderboard | |
| 2 | DeepLabV3 R-101-D8 | 43.4% | 320000 iterations, 512x512 crop, Multiscale and flip | Official leaderboard | |
| 3 | PSPNet R-101-D8 | 42.4% | 320000 iterations, 512x512 crop, Multiscale and flip | Official leaderboard | |
| 4 | DeepLabV3 R-50-D8 | 42.2% | 320000 iterations, 512x512 crop, Multiscale and flip | Official leaderboard | |
| 5 | SAN ViT-B_16 | 41.9% | 60000 iterations, 640x640 crop, CLIP_ViT-B16 pretraining, Single scale | Official leaderboard | |
| 6 | PSPNet R-50-D8 | 40.8% | 320000 iterations, 512x512 crop, Multiscale and flip | Official leaderboard | |
| 7 | BiSeNetV1 R-101-D32 | 38% | 160000 iterations, 512x512 crop, Multiscale and flip | Official leaderboard | |
| 8 | BiSeNetV1 R-50-D32 | 35.4% | 160000 iterations, 512x512 crop, Multiscale and flip | Official leaderboard | |
| 9 | BiSeNetV1 R-101-D32 (No Pretrain) | 31.8% | 160000 iterations, 512x512 crop, Multiscale and flip | Official leaderboard | |
| 10 | BiSeNetV1 R-50-D32 (No Pretrain) | 30.3% | 160000 iterations, 512x512 crop, Multiscale and flip | Official leaderboard | |
| 11 | BiSeNetV1 R-18-D32 | 29.3% | 160000 iterations, 512x512 crop, Multiscale and flip | Official leaderboard | |
| 12 | BiSeNetV1 R-18-D32 (No Pretrain) | 26.1% | 160000 iterations, 512x512 crop, Multiscale and flip | Official leaderboard |
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Vision
- MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy.
Qwen 3.6 Plus 86%
- MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. GPT-5.6 Sol 84.6%
- CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
- MathVista In use Math reasoning in visual contexts (charts, geometry, puzzles) from 31 datasets. Answer accuracy.
Seed2.1 Pro 90.7%
- MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy.
Qwen 3.8 Max 97.7%
- MMStar In use Models answer 1,500 human-curated image questions that cannot be solved without the image, across 6 core capabilities. Scored by accuracy.
Qwen 3.8 Max 85.9%
Data audit
All 8 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
- Pass Independent runs Scores from the official leaderboard (12 entries).
- Pass Models scored 12 entries on its leaderboard, none a model we track.
- Note Newest score No scores yet.
- Note Frontier models No models scored yet.
- Note Headroom No scores yet.
Leaderboard import 3 pass
- Pass Importer We read the board on MMSegmentation: 12 entries at the last read, none a model we track.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.