Surgical Phase Recognition on Cholec80 (Top-5 Accuracy)
70.35Top-5 Overall AccuracyCLIP-based multi-modal framework
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| CLIP-based multi-modal frameworkTraining Configuration=Cholec80 FT (Init: JIGSAWS)2026.03 | 70.35 | 65.83 | 75.84 | 51.31 | 74.07 | 61.95 | 44.41 | 65.33 | |
| CLIPTraining Configuration=Baseline2026.03 | 37.59 | 15.18 | 70.9 | 3.12 | 0.78 | 0.65 | 0 | 0 | |
| CLIPTraining Configuration=Cholec80 Only FT2026.03 | 26.46 | 99.2 | 21.23 | 79.43 | 14.58 | 5.62 | 6.09 | 0 | |
| CLIPTraining Configuration=JIGSAWS FT2026.03 | 3.95 | 99.9 | 0.2 | 0.88 | 0 | 0 | 0 | 0 |