Model card
OneFormer (DiNAT-L).
SHI Labsopen-sourceUnknown paramsUniversal image segmentation with task-conditioned joint training (DiNAT-L backbone)
First multi-task universal image segmentation framework. Task-conditioned architecture trained once on panoptic, instance, and semantic jointly. DiNAT-L backbone achieves 58.3 mIoU (ss) on ADE20K val. CVPR 2023.
§ 02 · Benchmarks
Every benchmark OneFormer (DiNAT-L) has a recorded score for.
| # | Benchmark | Area · Task | Metric | Value | Rank | Date | Source |
|---|---|---|---|---|---|---|---|
| 01 | Cityscapes | Vision & Documents · Semantic Segmentation | miou | 83.0% | #3 | 2026-04-20 | source ↗ |
| 02 | ADE20K | Vision & Documents · Semantic Segmentation | mIoU | 58.3% | #8 | 2026-04-20 |
Rank column shows this model’s position vs all other models scored on the same benchmark + metric (competitors after the slash). #1 in red means current SOTA. Sorted by rank, then newest result.
§ 03 · Strengths by area
Where OneFormer (DiNAT-L) actually performs.
§ 06 · Sources & freshness
Where these numbers come from.
codesota-api
1
result
src
1
result
1 of 2 rows marked verified.