Semantic Segmentation on COCO-Stuff C=27 (Held-out Seen/Unseen)
34.6mIoU (Seen)LILA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LILABackbone=ViT-B/14, Zero-shot setting=true, Linear probing=true, CLIP text embedding initialization=true2026.04 | 34.6 | 22.7 | 27.4 | |
| LILABackbone=ViT-S/14, Zero-shot setting=true, Linear probing=true, CLIP text embedding initialization=true2026.04 | 32.8 | 20.1 | 24.9 | |
| DINO2-B14Backbone=ViT-B/14, Zero-shot setting=true, Linear probing=true, CLIP text embedding initialization=true2026.04 | 32.6 | 19 | 24.1 | |
| DINO2-S14Backbone=ViT-S/14, Zero-shot setting=true, Linear probing=true, CLIP text embedding initialization=true2026.04 | 30.1 | 17 | 21.8 |