Surgical Phase Recognition on Cholec80
96AccuracySurgicalMamba
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| SurgicalMambaEvaluation protocol=Relaxed (10-second boundary)2026.05 | 96 | — | — | 88.5 | — | — | 94.9 | 94.4 | — | — | |
| DACATEvaluation protocol=Relaxed (10-second boundary)2026.05 | 95.5 | — | — | 87.4 | — | — | 93.6 | 93.4 | — | — | |
| MTTR-NetEvaluation protocol=Relaxed (10-second boundary)2026.05 | 95 | — | — | 84.7 | — | — | 92.7 | 91.2 | — | — | |
| SurgicalMambaEvaluation protocol=Strict (unrelaxed)2026.05 | 94.6 | — | — | 82.7 | — | — | 89.6 | 90.5 | — | — | |
| MTTR-NetEvaluation protocol=Strict (unrelaxed)2026.05 | 93.9 | — | — | 80.5 | — | — | 88.8 | 88.2 | — | — | |
| SKiTEvaluation protocol=Relaxed (10-second boundary)2026.05 | 93.4 | — | — | 82.6 | — | — | 90.9 | 91.8 | — | — | |
| SurgformerEvaluation protocol=Relaxed (10-second boundary)2026.05 | 93.4 | — | — | 84.1 | — | — | 91.9 | 92.1 | — | — | |
| CMTNetEvaluation protocol=Relaxed (10-second boundary)2026.05 | 92.9 | — | — | 81.5 | — | — | 90.1 | 92 | — | — | |
| SR-MambaEvaluation protocol=Relaxed (10-second boundary)2026.05 | 92.6 | — | — | 81.5 | — | — | 90.3 | 90.6 | — | — | |
| SKiTEvaluation protocol=Strict (unrelaxed)2026.05 | 92.5 | — | — | 76.7 | — | — | 84.6 | 88.5 | — | — | |
| LoViTEvaluation protocol=Relaxed (10-second boundary)2026.05 | 92.4 | — | — | 81.2 | — | — | 89.9 | 90.6 | — | — | |
| SurgformerEvaluation protocol=Strict (unrelaxed)2026.05 | 92.4 | — | — | 79.9 | — | — | 87.9 | 89.3 | — | — | |
| UATDEvaluation protocol=Relaxed (10-second boundary)2026.05 | 91.9 | — | — | 79.9 | — | — | 89.5 | 90.5 | — | — | |
| Not E2EEvaluation protocol=Relaxed (10-second boundary)2026.05 | 91.5 | — | — | 77.2 | — | — | — | 86.8 | — | — | |
| LoViTEvaluation protocol=Strict (unrelaxed)2026.05 | 91.5 | — | — | 74.2 | — | — | 83.1 | 86.5 | — | — | |
| OperaEvaluation protocol=Relaxed (10-second boundary)2026.05 | 91.2 | — | — | — | — | — | 82.2 | 86.9 | — | — | |
| Trans-SVNetEvaluation protocol=Relaxed (10-second boundary)2026.05 | 90.3 | — | — | 79.3 | — | — | 90.7 | 88.8 | — | — | |
| TMRNetEvaluation protocol=Relaxed (10-second boundary)2026.05 | 90.1 | — | — | 79.1 | — | — | 90.3 | 89.5 | — | — | |
| MTRCNet-CLEvaluation protocol=Relaxed (10-second boundary)2026.05 | 89.2 | — | — | — | — | — | 86.9 | 88 | — | — | |
| Trans-SVNetEvaluation protocol=Strict (unrelaxed)2026.05 | 89.1 | — | — | 72.5 | — | — | 84.7 | 83.6 | — | — | |
| TeCNOEvaluation protocol=Relaxed (10-second boundary)2026.05 | 88.6 | — | — | 75.1 | — | — | 81.6 | 85.2 | — | — | |
| SV-RCNetEvaluation protocol=Relaxed (10-second boundary)2026.05 | 85.3 | — | — | — | — | — | 80.7 | 83.5 | — | — | |
| PL-StitchPretraining data=LEMON, Backbone=ViT-B/16, Protocol=k-NN2025.11 | 81.7 | — | — | — | — | — | — | — | — | — | |
| EndoNetEvaluation protocol=Relaxed (10-second boundary)2026.05 | 81.7 | — | — | — | — | — | 73.7 | 79.6 | — | — | |
| SurgLIMEType=VL, Backbone=ViT-B/16, Evaluation Protocol=Linear probing2026.04 | 80.6 | — | — | — | — | — | — | — | — | 73.2 | |
| PL-StitchPretraining data=LEMON, Backbone=ViT-B/16, Protocol=Linear probing2025.11 | 80.4 | — | — | — | — | — | — | — | — | 73 | |
| PL-StitchType=G, Backbone=ViT-B/16, Evaluation Protocol=Linear probing2026.04 | 80.4 | — | — | — | — | — | — | — | — | 73 | |
| PhaseNetEvaluation protocol=Relaxed (10-second boundary)2026.05 | 78.8 | — | — | — | — | — | 71.3 | 76.6 | — | — | |
| LemonFMPretraining data=LEMON, Evaluation protocol=linear probing2025.03 | 75.8 | — | — | — | — | — | — | — | — | 68.6 | |
| iBOTType=G, Pretraining data=LEMON, Backbone=ViT-B/16, Protocol=Linear probing2025.11 | 74.6 | — | — | — | — | — | — | — | — | 67.6 | |
| iBOTType=G, Backbone=ViT-B/16, Evaluation Protocol=Linear probing2026.04 | 74.6 | — | — | — | — | — | — | — | — | 67.6 | |
| LemonFMPretraining data=LEMON, Backbone=ConvNeXt-B, Protocol=Linear probing2025.11 | 73.9 | — | — | — | — | — | — | — | — | 65.8 | |
| LemonFMType=S, Backbone=ConvNeXt-B, Evaluation Protocol=Linear probing2026.04 | 73.9 | — | — | — | — | — | — | — | — | 65.8 | |
| MocoPre-training Dataset=Cholec80, Shot=100, Linear probing=true2026.06 | 73.4 | — | — | — | — | — | — | — | — | 62.8 | |
| SurgeNetXLPretraining data=SurgeNetXL, Evaluation protocol=linear probing2025.03 | 73.2 | — | — | — | — | — | — | — | — | 65.1 | |
| DINOType=G, Pretraining data=LEMON, Backbone=ViT-B/16, Protocol=Linear probing2025.11 | 72.2 | — | — | — | — | — | — | — | — | 67.1 | |
| DINOType=G, Backbone=ViT-B/16, Evaluation Protocol=Linear probing2026.04 | 72.2 | — | — | — | — | — | — | — | — | 67.1 | |
| HyperVLPPre-training Dataset=SVL(VL), Shot=100, Linear probing=true2026.06 | 71.3 | — | — | — | — | — | — | — | — | 60.2 | |
| iBOTType=G, Pretraining data=LEMON, Backbone=ViT-B/16, Protocol=k-NN2025.11 | 70.3 | — | — | — | — | — | — | — | — | — | |
| PeskaVLPPre-training Dataset=SVL(VL), Shot=100, Linear probing=true2026.06 | 69.9 | — | — | — | — | — | — | — | — | 59.8 | |
| LemonFMPretraining data=LEMON, Backbone=ConvNeXt-B, Protocol=k-NN2025.11 | 69.5 | — | — | — | — | — | — | — | — | — | |
| DINOType=G, Pretraining data=LEMON, Backbone=ViT-B/16, Protocol=k-NN2025.11 | 69.4 | — | — | — | — | — | — | — | — | — | |
| MocoPre-training Dataset=SVL(V), Shot=100, Linear probing=true2026.06 | 68.2 | — | — | — | — | — | — | — | — | 55.8 | |
| ImageNetPre-training Dataset=ImageNet, Shot=100, Linear probing=true2026.06 | 66.4 | — | — | — | — | — | — | — | — | 54.9 | |
| HecVLPre-training Dataset=SVL(L), Shot=100, Linear probing=true2026.06 | 66 | — | — | — | — | — | — | — | — | 53.2 | |
| T-CoReType=G, Pretraining data=LEMON, Backbone=ViT-B/16, Protocol=Linear probing2025.11 | 65.1 | — | — | — | — | — | — | — | — | 59.4 | |
| CLIPPre-training Dataset=SVL(L), Shot=100, Linear probing=true2026.06 | 64.9 | — | — | — | — | — | — | — | — | 55 | |
| CLIPType=VL, Backbone=ViT-B/16, Evaluation Protocol=Linear probing2026.04 | 64.8 | — | — | — | — | — | — | — | — | 50.7 | |
| CLIPPre-training Dataset=NA(L), Shot=100, Linear probing=true2026.06 | 64.8 | — | — | — | — | — | — | — | — | 50.7 | |
| SurgVLPType=VL, Backbone=ResNet50, Evaluation Protocol=Linear probing2026.04 | 63.5 | — | — | — | — | — | — | — | — | 50.3 | |
| SurgVLPPre-training Dataset=SVL(L), Shot=100, Linear probing=true2026.06 | 63.5 | — | — | — | — | — | — | — | — | 50.3 | |
| Endo-FMPretraining data=Private data, Evaluation protocol=linear probing2025.03 | 62.7 | — | — | — | — | — | — | — | — | 53.9 | |
| Endo-FMType=S, Pretraining data=Private data, Backbone=ViT-B/16, Protocol=Linear probing2025.11 | 62.7 | — | — | — | — | — | — | — | — | 53.9 | |
| T-CoReType=G, Pretraining data=LEMON, Backbone=ViT-B/16, Protocol=k-NN2025.11 | 62.3 | — | — | — | — | — | — | — | — | — | |
| Endo-FMType=S, Pretraining data=Private data, Backbone=ViT-B/16, Protocol=k-NN2025.11 | 55.9 | — | — | — | — | — | — | — | — | — | |
| VideoMAEv2Pretraining data=LEMON, Evaluation protocol=linear probing2025.03 | 55.8 | — | — | — | — | — | — | — | — | 48.5 | |
| VideoMAEv2Type=G, Pretraining data=LEMON, Backbone=ViT-B/16, Protocol=Linear probing2025.11 | 55.8 | — | — | — | — | — | — | — | — | 48.5 | |
| VideoMAEv2Type=G, Backbone=ViT-B/16, Evaluation Protocol=Linear probing2026.04 | 55.8 | — | — | — | — | — | — | — | — | 48.5 | |
| MAEPretraining data=LEMON, Evaluation protocol=linear probing2025.03 | 54.9 | — | — | — | — | — | — | — | — | 43.4 | |
| MAEType=G, Pretraining data=LEMON, Backbone=ViT-B/16, Protocol=Linear probing2025.11 | 54.9 | — | — | — | — | — | — | — | — | 43.4 | |
| MAEType=G, Backbone=ViT-B/16, Evaluation Protocol=Linear probing2026.04 | 54.9 | — | — | — | — | — | — | — | — | 43.4 | |
| VideoMAEv2Type=G, Pretraining data=LEMON, Backbone=ViT-B/16, Protocol=k-NN2025.11 | 50 | — | — | — | — | — | — | — | — | — | |
| MAEType=G, Pretraining data=LEMON, Backbone=ViT-B/16, Protocol=k-NN2025.11 | 47.2 | — | — | — | — | — | — | — | — | — | |
| EndoViTPretraining data=Merged public data, Evaluation protocol=linear probing2025.03 | 46.9 | — | — | — | — | — | — | — | — | 39 | |
| EndoViTType=S, Pretraining data=Merged public data, Backbone=ViT-B/16, Protocol=Linear probing2025.11 | 46.9 | — | — | — | — | — | — | — | — | 39 | |
| EndoViTType=S, Pretraining data=Merged public data, Backbone=ViT-B/16, Protocol=k-NN2025.11 | 45.8 | — | — | — | — | — | — | — | — | — | |
| SurgVLPBackbone=ResNet50, Zero-shot=true2026.04 | 34.7 | — | — | — | — | — | — | — | — | 24.4 | |
| SurgLIMEBackbone=ViT-B/16, Zero-shot=true2026.04 | 33 | — | — | — | — | — | — | — | — | 8.7 | |
| CLIPBackbone=ViT-B/16, Zero-shot=true2026.04 | 27.8 | — | — | — | — | — | — | — | — | 8.4 | |
| GSViTPretraining data=GenSurgery, Evaluation protocol=linear probing2025.03 | 20.2 | — | — | — | — | — | — | — | — | 13.5 | |
| CLIPShots=02025.06 | — | 13.1 | — | — | — | — | — | — | — | — | |
| CLIPPretraining dataset=CLIP400M, Zero-shot=true2024.05 | — | 13.1 | 30.8 | — | — | — | — | — | — | — | |
| CLIP*Pretraining dataset=SVL-Pretrain, Zero-shot=true, Initialization=OpenAI2024.05 | — | 19.6 | 33.8 | — | — | — | — | — | — | — | |
| CLIP**Pretraining dataset=SVL-Pretrain, Zero-shot=true, Initialization=random2024.05 | — | 10.4 | 29.4 | — | — | — | — | — | — | — | |
| CliPPERZero-shot=true, Number of Prompts=4, Pre-training Data=SVL2026.03 | — | — | — | — | — | — | — | — | — | 30.6 | |
| CliPPERZero-shot=true, Number of Prompts=4, Pre-training Data=YouTube2026.03 | — | — | — | — | — | — | — | — | — | 34.9 | |
| CliPPERZero-shot=true, Number of Prompts=4, Pre-training Data=SVL + YouTube2026.03 | — | — | — | — | — | — | — | — | — | 38.3 | |
| Dino v3Scale=L2026.02 | — | 81.19 | 87.2 | 69.6 | — | — | — | — | — | — | |
| Dino v3Scale=H2026.02 | — | 78.64 | 86.25 | 68.84 | — | — | — | — | — | — | |
| DINOv3Pre-training Domain=Generic2026.03 | — | — | 22.34 | — | — | — | — | — | — | — | |
| DINOv3Pre-training Domain=Surgical2026.03 | — | — | 16.12 | — | — | — | — | — | — | — | |
| Endo SSL2026.02 | — | 38.73 | 58.7 | 28.07 | — | — | — | — | — | — | |
| Endo-FMType=F, Fine-tuning Protocol=Full2025.03 | — | — | 86.9 | 76.7 | — | — | — | — | — | — | |
| Endo-Mamba2026.02 | — | 54.69 | 68.81 | 42.58 | — | — | — | — | — | — | |
| EndoFM2026.02 | — | 36.13 | 57.25 | 26.42 | — | — | — | — | — | — | |
| EndoFMBackbone=ViT-B, Pre-training type=Existing pretrained, Backbone status=Frozen2026.02 | — | — | — | — | 62.91 | 76.47 | 72.1 | 65.79 | 53 | — | |
| EndoViT2026.02 | — | 36.78 | 56.41 | 26.92 | — | — | — | — | — | — | |
| EndoViTBackbone=ViT-B, Pre-training type=Existing pretrained, Backbone status=Frozen2026.02 | — | — | — | — | 63.08 | 78.6 | 73.77 | 66.71 | 53.36 | — | |
| EndoViTType=F, Fine-tuning Protocol=Full2025.03 | — | — | 86.7 | 75.3 | — | — | — | — | — | — | |
| GastroNet2026.02 | — | 81.22 | 88.9 | 72.02 | — | — | — | — | — | — | |
| Gemini-2.0-FlashThinking=false2026.03 | — | — | 38.89 | — | — | — | 40.03 | 36.76 | 21.47 | — | |
| Gemini-3.0-Pro-ThinkingThinking=true2026.03 | — | — | 52.35 | — | — | — | 43.08 | 44.8 | 29.59 | — | |
| Gemma3-27B-itThinking=false2026.03 | — | — | 14.08 | — | — | — | 19.67 | 17.49 | 4.45 | — | |
| GPT-4o-2024-0806Thinking=false2026.03 | — | — | 36.43 | — | — | — | 32.99 | 31.02 | 19.55 | — | |
| GPT-5.1-ThinkingThinking=true2026.03 | — | — | 49.7 | — | — | — | 38.33 | 38.84 | 24.95 | — | |
| GSViT2026.02 | — | 8.06 | 39.1 | 5.71 | — | — | — | — | — | — | |
| GSViTBackbone=EfficientViT, Pre-training type=Existing pretrained, Backbone status=Frozen2026.02 | — | — | — | — | 43.96 | 63.74 | 54.66 | 49.36 | 34.7 | — | |
| GSViTType=F, Fine-tuning Protocol=Full2025.03 | — | — | 75.3 | 57.8 | — | — | — | — | — | — | |
| HecVLPretraining dataset=SVL-Pretrain, Zero-shot=true2024.05 | — | 26.3 | 41.7 | — | — | — | — | — | — | — | |
| HecVLZero-shot=true, Number of Prompts=4, Pre-training Data=SVL2026.03 | — | — | — | — | — | — | — | — | — | 22.5 |