Model card
ViT-Adapter-L (BEiT-3).
Microsoft Researchopen-sourceUnknown paramsViT-L with spatial prior adapter + BEiT-3 pre-training + Mask2Former head
ViT-Adapter bridges the gap between plain ViT and hierarchical backbones for dense predictions. BEiT-3 pre-trained ViT-L via ViT-Adapter sets SOTA of 62.8 MS mIoU on ADE20K val. ICLR 2023 Spotlight.
§ 02 · Benchmarks
Every benchmark ViT-Adapter-L (BEiT-3) has a recorded score for.
| # | Benchmark | Area · Task | Metric | Value | Rank | Date | Source |
|---|---|---|---|---|---|---|---|
| 01 | ADE20K | Vision & Documents · Semantic Segmentation | mIoU | 62.8% | #3 | 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 ViT-Adapter-L (BEiT-3) actually performs.
§ 05 · Related models
Other Microsoft Research models scored on Codesota.
§ 06 · Sources & freshness
Where these numbers come from.
src
1
result
0 of 1 rows marked verified.