Keyframe detection on Blind-sweep US dataset
64.2PrecisionSA+MS+MT+PRS+L
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SA+MS+MT+PRS+LStructure-augmented segmentation prior=True, Multi-depth spatial encoding=True, Local and global temporal conditioning=True, Posterior refinement smoothing=True, Full objective (+L)=True2026.04 | 64.2 | 70.07 | 67.01 | 15.33 | 10.04 | |
| SA+MS+MT+PRSStructure-augmented segmentation prior=True, Multi-depth spatial encoding=True, Local and global temporal conditioning=True, Posterior refinement smoothing=True2026.04 | 61.11 | 72.63 | 66.37 | 16.45 | 10.82 | |
| SA+MS+MTStructure-augmented segmentation prior=True, Multi-depth spatial encoding=True, Local and global temporal conditioning=True2026.04 | 60.14 | 72.63 | 65.8 | 16.82 | 10.91 | |
| nnU-Net2026.04 | 52.65 | 63.58 | 57.6 | 77.78 | 10.07 | |
| MTLocal and global temporal conditioning=True2026.04 | 46.55 | 41 | 43.6 | 49.29 | 11.29 | |
| MS+MTMulti-depth spatial encoding=True, Local and global temporal conditioning=True2026.04 | 43.73 | 47.92 | 45.73 | 52.59 | 11.2 | |
| MSMulti-depth spatial encoding=True2026.04 | 40.21 | 45.47 | 42.68 | 60.57 | 11.8 | |
| BiomedCLIPBackbone=ViT-B2026.04 | 37.83 | 51.12 | 43.48 | 46.47 | 14 |