Action Triplet Recognition on CholecT50
91.55AP (I)UniSurg
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| UniSurg2026.02 | 91.55 | 57.72 | 48.18 | 40.39 | 43.47 | 39.54 | |
| Dino v3-HModel size=Huge2026.02 | 90.18 | 55.83 | 42.7 | 36.96 | 34.09 | 30.06 | |
| Dino v3-LModel size=Large2026.02 | 87.16 | 53.9 | 43.44 | 37.31 | 40.81 | 36.85 | |
| ZENBackbone=ViT-B, Pre-training Strategy=Existing pretrained, Evaluation Protocol=Fine-tuned2026.02 | 86.11 | 58.98 | 46.07 | 35.15 | 40.15 | 32.07 | |
| MIS-DINOv2Backbone=ViT-L, Pre-training Strategy=Self-supervised learning, Evaluation Protocol=Fine-tuned2026.02 | 85.4 | 56.97 | 43.26 | 33.52 | 38 | 29.37 | |
| MIS-DINOBackbone=ViT-B, Pre-training Strategy=Self-supervised learning, Evaluation Protocol=Fine-tuned2026.02 | 83.61 | 55.59 | 39.7 | 31.51 | 33.98 | 26.63 | |
| MIS-MSNBackbone=ViT-B, Pre-training Strategy=Self-supervised learning, Evaluation Protocol=Fine-tuned2026.02 | 83.54 | 55.34 | 40 | 32.03 | 35.56 | 27.89 | |
| SurgeNetBackbone=CAFormer18, Pre-training Strategy=Existing pretrained, Evaluation Protocol=Fine-tuned2026.02 | 83.22 | 54.61 | 39.8 | 31.73 | 33.83 | 26.91 | |
| MIS-DINOv2Backbone=ViT-B, Pre-training Strategy=Self-supervised learning, Evaluation Protocol=Fine-tuned2026.02 | 82.8 | 55.23 | 42.43 | 32.16 | 36.35 | 28.13 | |
| VideoMAE2026.02 | 81.58 | 47.93 | 36.78 | 31.11 | 30.4 | 25.72 | |
| PeskaVLPBackbone=ResNet50, Pre-training Strategy=Existing pretrained, Evaluation Protocol=Fine-tuned2026.02 | 81.24 | 54.58 | 38.58 | 31.52 | 34.19 | 26.04 | |
| ImageNetBackbone=ViT-B, Pre-training Strategy=Existing pretrained, Evaluation Protocol=Fine-tuned2026.02 | 80.91 | 52.84 | 39.4 | 30.26 | 34.09 | 26.57 | |
| MIS-MoCoV3Backbone=ViT-B, Pre-training Strategy=Self-supervised learning, Evaluation Protocol=Fine-tuned2026.02 | 79.2 | 50.92 | 36.97 | 29.59 | 32.23 | 24.97 | |
| EndoFMBackbone=ViT-B, Pre-training Strategy=Existing pretrained, Evaluation Protocol=Fine-tuned2026.02 | 78.18 | 51.06 | 37.39 | 28.97 | 32.77 | 25.2 | |
| VideoMAE-GModel size=Giant2026.02 | 77.06 | 46.69 | 40.42 | 30.6 | 29.08 | 24.41 | |
| SurgVLP2026.02 | 75.56 | 44.75 | 33.29 | 29.3 | 28.26 | 24.11 | |
| MIS-MAEBackbone=ViT-B, Pre-training Strategy=Self-supervised learning, Evaluation Protocol=Fine-tuned2026.02 | 74.86 | 48.97 | 37.6 | 27.71 | 31.44 | 24.53 | |
| SurgNetXL2026.02 | 74.64 | 45.58 | 40.08 | 29.92 | 32.51 | 27.36 | |
| GastroNet2026.02 | 73.69 | 48.14 | 39.44 | 31.22 | 30.05 | 27.37 | |
| EndoViTBackbone=ViT-B, Pre-training Strategy=Existing pretrained, Evaluation Protocol=Fine-tuned2026.02 | 73.24 | 48.63 | 36.58 | 27.3 | 31.19 | 24.49 | |
| Endo Mamba2026.02 | 58.45 | 36.23 | 31.55 | 22.86 | 23.54 | 20.05 | |
| EndoViT2026.02 | 52.41 | 32.01 | 27.45 | 20.65 | 19.42 | 18.05 | |
| EndoSSL2026.02 | 50.47 | 34.79 | 27.93 | 22.67 | 20.2 | 20.17 | |
| EndoFM2026.02 | 48.57 | 28.52 | 27.51 | 18.44 | 19.71 | 18.98 | |
| Selfsupsurg2026.02 | 45.5 | 29.16 | 22.65 | 18.25 | 14.41 | 13.5 | |
| GSViTBackbone=EfficientViT, Pre-training Strategy=Existing pretrained, Evaluation Protocol=Fine-tuned2026.02 | 32.27 | 21.12 | 15.91 | 11.08 | 8.43 | 6.76 | |
| GSViT2026.02 | 18.65 | 12.17 | 8.4 | 6.36 | 3.56 | 3.2 |