Few-Shot Semantic Segmentation on COCO-20i
75.8mIoUFSS-SAM3
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| FSS-SAM3Shots=1-shot, Textual Priors=True2026.04 | 75.8 | 74 | 75 | 76.7 | 77.5 | 87.1 | |
| FSS-SAM3Shots=5-shot, Textual Priors=True2026.04 | 75.1 | 74.5 | 73.2 | 76 | 76.8 | 86.4 | |
| FSS-SAM3Shots=5-shot, Textual Priors=False2026.04 | 71.8 | 68.6 | 72.5 | 72.5 | 73.6 | 84.9 | |
| FSSAMShots=5-shot2026.04 | 69.3 | 68.6 | 74 | 64.5 | 69.9 | 82.9 | |
| SANSAShots=5-shot2026.04 | 67.3 | — | — | — | — | — | |
| GF-SAMShots=5, In-domain training=false, Type=generalist model2024.10 | 66.8 | — | — | — | — | — | |
| GF-SAMShots=5-shot2026.04 | 66.8 | 67.1 | 69.4 | 66 | 64.8 | — | |
| FSS-SAM3Shots=1-shot, Textual Priors=False2026.04 | 66.6 | 64.2 | 66.8 | 67.7 | 67.5 | 82.9 | |
| FSSAMShots=1-shot2026.04 | 62.3 | 59.9 | 65.6 | 62.1 | 61.6 | 77.3 | |
| MatcherShots=5, In-domain training=false, Type=generalist model2024.10 | 60.7 | — | — | — | — | — | |
| MatcherShots=5-shot2026.04 | 60.7 | 60.1 | 62.7 | 60.9 | 59.2 | — | |
| SANSAShots=1-shot2026.04 | 60.2 | 58.9 | 62.6 | 61.5 | 58 | — | |
| VINEIE=ResNet, Shots=5-shot2026.03 | 59.3 | 54.1 | 64.6 | 61.3 | 57.4 | — | |
| HMNetIE=ResNet, Shots=5-shot2026.03 | 58.9 | 53.4 | 64.6 | 60.8 | 56.8 | — | |
| FPTransShots=5-shot2026.04 | 58.9 | 54.2 | 62.5 | 61.3 | 57.6 | — | |
| HMNetShots=5-shot2026.04 | 58.9 | 53.4 | 64.6 | 60.8 | 56.8 | 77.6 | |
| GF-SAMShots=1, In-domain training=false, Type=generalist model2024.10 | 58.7 | — | — | — | — | — | |
| GF-SAMShots=1-shot2026.04 | 58.7 | 56.6 | 61.4 | 59.6 | 57.1 | — | |
| DCAMABackbone=Swin-B, Type=Pixel-wise, Shots=5-shot2022.07 | 58.3 | 55.4 | 60.3 | 59.9 | 57.5 | 76.9 | |
| FCP†IE=ResNet, Shots=5-shot2026.03 | 58 | 52.6 | 63.3 | 59.8 | 56.1 | — | |
| DGPNetBackbone=ResNet101, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 57.9 | — | — | — | — | — | |
| AMFormerShots=5, In-domain training=true, Type=specialist model2024.10 | 57.3 | — | — | — | — | — | |
| AMNetShots=5-shot2026.04 | 57.3 | 52 | 61.9 | 57.4 | 57.9 | 78.8 | |
| AENetShots=5-shot2026.04 | 56.7 | 51.7 | 61.9 | 57.9 | 55.3 | 76.5 | |
| DGPNetBackbone=ResNet50, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 56.2 | — | — | — | — | — | |
| HDMNetShots=5, In-domain training=true, Type=specialist model2024.10 | 56 | — | — | — | — | — | |
| HDMNetIE=ResNet, Shots=5-shot2026.03 | 56 | 50.6 | 61.6 | 55.7 | 56 | — | |
| HDMNetShots=5-shot2026.04 | 56 | 50.6 | 61.6 | 55.7 | 56 | 77.7 | |
| VRP-SAM†IE=ResNet, Shots=5-shot2026.03 | 55.5 | 50.5 | 59.5 | 56.9 | 54.9 | — | |
| HSNetBackbone=Swin-B, Type=Pixel-wise, Shots=5-shot, is_reimplementation=true2022.07 | 55.1 | 50.1 | 58.6 | 56.7 | 55.1 | 76.1 | |
| HMNetIE=VGG, Shots=5-shot2026.03 | 54.5 | 48.8 | 58 | 57.9 | 53.2 | — | |
| VINEIE=VGG, Shots=5-shot2026.03 | 54.5 | 49.8 | 59.4 | 56.1 | 52.7 | — | |
| UniFSSIE=ResNet, Shots=5-shot2026.03 | 54.1 | 50.3 | 59.5 | 54.4 | 52 | — | |
| VRP-SAMShots=1, In-domain training=false, Type=generalist model2024.10 | 53.9 | — | — | — | — | — | |
| SCCANShots=5-shot2026.04 | 53.9 | 47.2 | 57.2 | 59.2 | 52.1 | 74.2 | |
| PAMShots=5-shot2026.04 | 53.9 | 48.1 | 60.8 | 54.8 | 51.9 | — | |
| VRP-SAMShots=1-shot2026.04 | 53.9 | 48.1 | 55.8 | 60 | 51.6 | — | |
| FCP†IE=VGG, Shots=5-shot2026.03 | 53.7 | 49 | 58.3 | 55.2 | 52.3 | — | |
| VINEIE=ResNet, Shots=1-shot2026.03 | 53.7 | 47.7 | 57.4 | 56.7 | 52.9 | — | |
| MatcherShots=1, In-domain training=false, Type=generalist model2024.10 | 52.7 | — | — | — | — | — | |
| MatcherShots=1-shot2026.04 | 52.7 | 52.7 | 53.5 | 52.6 | 52.1 | — | |
| FCP†IE=ResNet, Shots=1-shot2026.03 | 52.5 | 46.4 | 56.4 | 55.3 | 51.8 | — | |
| HDMNetIE=VGG, Shots=5-shot2026.03 | 52.4 | 47 | 56.5 | 54.1 | 51.9 | — | |
| HMNetIE=ResNet, Shots=1-shot2026.03 | 52.1 | 45.5 | 58.7 | 52.9 | 51.4 | — | |
| HMNetShots=1-shot2026.04 | 52.1 | 45.5 | 58.7 | 52.9 | 51.4 | 74.5 | |
| DCAMABackbone=ResNet-101, Type=Pixel-wise, Shots=5-shot2022.07 | 51.9 | 48 | 58 | 54.3 | 47.1 | 73.3 | |
| VRP-SAM†IE=VGG, Shots=5-shot2026.03 | 51.9 | 43.6 | 57.7 | 54.7 | 51.8 | — | |
| BAMShots=5-shot2026.04 | 51.2 | 49.3 | 54.2 | 51.6 | 49.6 | — | |
| AMFormerShots=1, In-domain training=true, Type=specialist model2024.10 | 51 | — | — | — | — | — | |
| VINEIE=VGG, Shots=1-shot2026.03 | 51 | 45.2 | 55.4 | 54.9 | 48.6 | — | |
| AMNetShots=1-shot2026.04 | 51 | 44.9 | 55.8 | 52.7 | 50.6 | 72.9 | |
| DCAMABackbone=Swin-B, Type=Pixel-wise, Shots=1-shot2022.07 | 50.9 | 49.5 | 52.7 | 52.8 | 48.7 | 73.2 | |
| VRP-SAM†IE=ResNet, Shots=1-shot2026.03 | 50.2 | 44.3 | 54.3 | 52.3 | 50 | — | |
| HDMNetShots=1, In-domain training=true, Type=specialist model2024.10 | 50 | — | — | — | — | — | |
| FCP†IE=VGG, Shots=1-shot2026.03 | 50 | 44.5 | 54.3 | 53.5 | 47.7 | — | |
| HDMNetIE=ResNet, Shots=1-shot2026.03 | 50 | 43.8 | 55.3 | 51.6 | 49.4 | — | |
| HDMNetShots=1-shot2026.04 | 50 | 43.8 | 55.3 | 51.6 | 49.4 | 72.2 | |
| HSNetBackbone=ResNet-101, Type=Pixel-wise, Shots=5-shot2022.07 | 49.5 | 45.9 | 53 | 51.8 | 47.1 | 72.4 | |
| HSNetBackbone=ResNet101, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 49.5 | — | — | — | — | — | |
| HSNetShots=5, In-domain training=true, Type=specialist model2024.10 | 49.5 | — | — | — | — | — | |
| AENetShots=1-shot2026.04 | 49.4 | 43.1 | 56 | 50.3 | 48.4 | 73.6 | |
| HMNetIE=VGG, Shots=1-shot2026.03 | 49.1 | 44.2 | 51.8 | 51.9 | 48.4 | — | |
| UniFSSIE=ResNet, Shots=1-shot2026.03 | 48.9 | 46.5 | 53 | 48 | 48.2 | — | |
| BAMIE=ResNet, Shots=5-shot2026.03 | 48.5 | 43.2 | 53.4 | 49.4 | 48.1 | — | |
| PAMShots=1-shot2026.04 | 48.5 | 44.1 | 55 | 46.5 | 48.5 | — | |
| DCAMABackbone=ResNet-50, Type=Pixel-wise, Shots=5-shot2022.07 | 48.3 | 45.9 | 50.5 | 50.7 | 46 | 71.7 | |
| VATShots=5, In-domain training=true, Type=specialist model2024.10 | 47.9 | — | — | — | — | — | |
| VRP-SAM†IE=VGG, Shots=1-shot2026.03 | 47.4 | 39.4 | 52 | 50.6 | 47.5 | — | |
| HSNetBackbone=Swin-B, Type=Pixel-wise, Shots=1-shot, is_reimplementation=true2022.07 | 47.3 | 43.6 | 49.9 | 49.4 | 46.4 | 72.5 | |
| FPTransShots=1-shot2026.04 | 47 | 44.4 | 48.9 | 50.6 | 44 | — | |
| HSNetBackbone=ResNet50, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 46.9 | — | — | — | — | — | |
| HSNetShots=5-shot2026.04 | 46.9 | 43.3 | 51.3 | 48.2 | 45 | — | |
| DGPNetBackbone=ResNet101, Shot=1-shot, Resolution=512x512, Number of episodes=20k2021.10 | 46.7 | — | — | — | — | — | |
| SCCANShots=1-shot2026.04 | 46.3 | 40.4 | 49.7 | 49.6 | 45.6 | 69.9 | |
| BAMShots=1-shot2026.04 | 46.2 | 43.4 | 50.6 | 47.5 | 43.4 | — | |
| HDMNetIE=VGG, Shots=1-shot2026.03 | 45.9 | 40.7 | 50.6 | 48.2 | 44 | — | |
| SCLBackbone=ResNet-50, Type=Prototype, Shots=5-shot2022.07 | 45.6 | 41.1 | 48.9 | 45.2 | 47 | — | |
| CyCTRBackbone=ResNet50, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 45.6 | — | — | — | — | — | |
| CyCTRIE=ResNet, Shots=5-shot2026.03 | 45.6 | 41.1 | 48.9 | 45.2 | 47 | — | |
| BAMIE=ResNet, Shots=1-shot2026.03 | 45.2 | 39.4 | 49.9 | 46.2 | 45.2 | — | |
| DGPNetBackbone=ResNet50, Shot=1-shot, Resolution=512x512, Number of episodes=20k2021.10 | 45 | — | — | — | — | — | |
| MLCBackbone=ResNet101, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 44.4 | — | — | — | — | — | |
| IMR-HSNetShot=Five, Backbone=Res502023.03 | 44.4 | 40.7 | 46 | 45 | 46 | — | |
| SSPIE=ResNet, Shots=5-shot2026.03 | 44.1 | 40 | 47 | 45.1 | 43.9 | — | |
| DCAMABackbone=ResNet-101, Type=Pixel-wise, Shots=1-shot2022.07 | 43.5 | 41.5 | 46.2 | 45.2 | 41.3 | 69.9 | |
| DCAMABackbone=ResNet-50, Type=Pixel-wise, Shots=1-shot2022.07 | 43.3 | 41.9 | 45.1 | 44.4 | 41.7 | 69.5 | |
| CMNBackbone=ResNet50, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 43.1 | — | — | — | — | — | |
| DENetBackbone=ResNet50, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 43 | — | — | — | — | — | |
| DENetBackbone=ResNet50, Shot=1-shot, Resolution=512x512, Number of episodes=20k2021.10 | 42.8 | — | — | — | — | — | |
| PEFNetBackbone=ResNet-101, Type=Prototype, Shots=5-shot2022.07 | 42.7 | 40.4 | 46.8 | 43.2 | 40.5 | 65.8 | |
| PFENetBackbone=ResNet101, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 42.7 | — | — | — | — | — | |
| SAGNNBackbone=ResNet101, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 42.7 | — | — | — | — | — | |
| ASGNetBackbone=ResNet101, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 42.5 | — | — | — | — | — | |
| IMR-HSNetShot=One, Backbone=Res502023.03 | 42.4 | 39.5 | 43.8 | 42.4 | 44.1 | — | |
| RePRIBackbone=ResNet-50, Type=Prototype, Shots=5-shot2022.07 | 42.1 | 38.5 | 46.2 | 40 | 43.6 | — | |
| RePriBackbone=ResNet50, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 42.1 | — | — | — | — | — | |
| CWTBackbone=ResNet-101, Type=Prototype, Shots=5-shot2022.07 | 42 | 38.5 | 46.7 | 39.4 | 43.2 | — | |
| CANetBackbone=ResNet50, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 42 | — | — | — | — | — | |
| CWTBackbone=ResNet101, Shot=5-shot, Resolution=512x512, Number of episodes=20k2021.10 | 42 | — | — | — | — | — | |
| MFNet + SABackbone=ResNet-50, Way=1-way, Shots=5-shot2021.10 | 41.9 | — | — | — | — | — |