Medical Image Segmentation on IMed-361M (test)
89.6DSC (Bbox)SyncSAM-IMed-50
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SyncSAM-IMed-50Backbone scale=502024.08 | 89.6 | 87.3 | 89.7 | |
| SyncSAM-IMed-34Backbone scale=342024.08 | 88.2 | 86.9 | 89.1 | |
| MedSAMTraining protocol=Fully fine-tuned2024.08 | 83.8 | — | — | |
| SAM-Med2DTraining protocol=Adapter-tuned2024.08 | 82.3 | 78.7 | 83.6 | |
| FT-SAMTraining protocol=Mask decoder trained2024.08 | 80.1 | 78.3 | 81.9 | |
| SAMProtocol=Direct inference2024.08 | 67.2 | 23.3 | 51.4 |