Instance Segmentation on SA-1B
36.2AR (Small)UnSAM+
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| UnSAM+Supervision Protocol=lightly semi-supervised, Backbone=RN-50, Number of parameters=23M, Training data size (Number of images)=0.1M2024.06 | 36.2 | — | 65.9 | 76.5 | 42.8 | 64.8 | |
| SAMSupervision=Supervised2024.02 | 20 | 60.8 | 59.9 | 82.2 | 38.9 | — | |
| SAMSupervision Protocol=fully-supervised, Backbone=ViT-B/8, Number of parameters=85M, Training data size (Number of images)=11M2024.06 | 20 | — | 59.9 | 82.8 | 38.9 | 60.8 | |
| HASSODSupervision=Self-supervised2024.02 | 12.9 | 26 | 22.8 | 38.3 | 13.8 | — | |
| CutLERSupervision=Self-supervised2024.02 | 4.9 | 17 | 13.9 | 28.5 | 7.8 | — | |
| FreeSOLOSupervision=Self-supervised2024.02 | 0 | 2.2 | 0.2 | 6.9 | 1.5 | — |