Semantic Segmentation on ADE20K (Novel and Zero-Shot mIoU)
38mIoU (novel)MaskCLIP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MaskCLIPMechanism=CLIP-based pseudo-labels, Notes=Improves with self-training2026.04 | 38 | — | |
| Semantic-Fast-SAM (Ours)Mechanism=FastSAM masks + BLIP/CLIP fusion + closed-set priors, Notes=High-detail masks and strong overall performance2026.04 | — | 53.7 |