Semantic Segmentation on Chest X-ray
87.5mIoUSegGPT
Evaluation results
| Method | Links | |
|---|---|---|
| SegGPTEncoder=ViT, #Param=354 M, Training Protocol=Task-specific fine-tuning, Supervision Type=Semantic + mask supervision2026.03 | 87.5 | |
| DMTNet (TaP)Version=TaP, Shot count=15-shot2025.12 | 86.76 | |
| DMTNet (TaP)Version=TaP, Shot count=10-shot2025.12 | 85.38 | |
| DMTNet (TaP)Version=TaP, Shot count=5-shot2025.12 | 81.53 | |
| INSID3Encoder=DINOv3, #Param=304 M, Training Protocol=Training free, Supervision Type=Unsupervised pre-training2026.03 | 78.8 | |
| DMTNet (TaP)Version=TaP, Shot count=3-shot2025.12 | 78.59 | |
| DMTNet (Decoder FT)Version=Decoder FT, Shot count=15-shot2025.12 | 76.48 | |
| DMTNet (Decoder FT)Version=Decoder FT, Shot count=10-shot2025.12 | 75.06 | |
| MatcherEncoder=DINOv2 + SAM, #Param=945 M, Training Protocol=Training free, Supervision Type=Mask-supervised pre-training2026.03 | 70.8 | |
| DMTNet (Decoder FT)Version=Decoder FT, Shot count=5-shot2025.12 | 70.79 | |
| DMTNet (Decoder FT)Version=Decoder FT, Shot count=3-shot2025.12 | 68.19 | |
| FRONTBackbone=ResNet-50, Few-shot protocol=5-shot, Source Domain=ImageNet-1K2026.03 | 67.29 | |
| DMTNet (Vanilla)Version=Vanilla, Shot count=10-shot2025.12 | 67.1 | |
| DMTNet (Vanilla)Version=Vanilla, Shot count=15-shot2025.12 | 66.11 | |
| DMTNet (Vanilla)Version=Vanilla, Shot count=5-shot2025.12 | 65.58 | |
| DMTNet (Vanilla)Version=Vanilla, Shot count=3-shot2025.12 | 64.83 | |
| GF-SAM† + debiasEncoder=DINOv3 + SAM, #Param=945 M, Training Protocol=Training free, Supervision Type=Mask-supervised pre-training, Debias=True2026.03 | 60 | |
| GF-SAM†Encoder=DINOv3 + SAM, #Param=945 M, Training Protocol=Training free, Supervision Type=Mask-supervised pre-training2026.03 | 56.1 | |
| GF-SAMEncoder=DINOv2 + SAM, #Param=945 M, Training Protocol=Training free, Supervision Type=Mask-supervised pre-training2026.03 | 51 | |
| DiffewSEncoder=Stable Diffusion, #Param=890 M, Training Protocol=Task-specific fine-tuning, Supervision Type=Semantic + mask supervision2026.03 | 41.6 | |
| SINEEncoder=DINOv2, #Param=373 M, Training Protocol=Task-specific fine-tuning, Supervision Type=Semantic + mask supervision2026.03 | 39.8 | |
| ScratchBackbone=ResNet-50, Few-shot protocol=5-shot, Source Domain=ImageNet-1K2026.03 | 38.56 | |
| SegICEncoder=DINOv2, #Param=310 M, Training Protocol=Task-specific fine-tuning, Supervision Type=Semantic + mask supervision2026.03 | 34.5 | |
| PerSAMEncoder=SAM, #Param=640 M, Training Protocol=Training free, Supervision Type=Mask-supervised pre-training2026.03 | 31.7 | |
| SegIC (COCO)Encoder=DINOv2, #Param=310 M, Training Protocol=Task-specific fine-tuning, Supervision Type=Semantic + mask supervision, dataset split=COCO2026.03 | 30.8 |