Table 3 Quantitative comparisons of WSFS+ with SAM-Med2D method and the combinations of WSFS/WSFS+ with SAM-Med2D where SAM-Med2D leverages the output of WSFS/WSFS+ in forms of point/bounding box/mask as prompt on USOVA3D dataset.
From: A weakly-supervised follicle segmentation method in ultrasound images
Method | mAP50 | IOU | Dice Score | FLOPs(G) | Params(M) |
|---|---|---|---|---|---|
WSFS+ | 0.957 | 0.714 | 0.83 | 10.5 | 71.8 |
SAM-Med2D | 0.716 | 0.373 | 0.54 | 33.8 | 271.2 |
WSFS(pt) & SAM-Med2D | 0.716 | 0.590 | 0.74 | 44.3 | 343.0 |
WSFS(box) & SAM-Med2D | 0.938 | 0.700 | 0.82 | 44.3 | 343.0 |
WSFS(mask) & SAM-Med2D | 0.921 | 0.703 | 0.83 | 44.3 | 343.0 |
WSFS+(pt) & SAM-Med2D | 0.795 | 0.596 | 0.75 | 44.3 | 343.0 |
WSFS+(box) & SAM-Med2D | 0.942 | 0.707 | 0.83 | 44.3 | 343.0 |
WSFS+(mask) & SAM-Med2D | 0.967 | 0.724 | 0.84 | 44.3 | 343.0 |
