Semantic Segmentation on COCO-Stuff 27
52mIoUSPARK
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SPARKzero-shot=true, training-free=true, features=diffusion2026.01 | 52 | — | |
| Seg4Diffzero-shot=true, training-free=true, features=diffusion2026.01 | 49.7 | — | |
| DiffCutzero-shot=true, training-free=true, features=diffusion2026.01 | 49.1 | — | |
| iSegzero-shot=true, training-free=true, features=diffusion2026.01 | 45.2 | — | |
| DINO-LEncoder=DINO-L, Model=Teacher (frozen DINO), Repr.=teacher, Evaluation Protocol=Linear probe2026.05 | 44.9 | — | |
| DINO-BEncoder=DINO-B, Model=Teacher (frozen DINO), Repr.=teacher, Evaluation Protocol=Linear probe2026.05 | 44.6 | — | |
| DINO-BRepresentation=teacher, Evaluation protocol=Linear probing2026.05 | 44.6 | — | |
| DiffSegzero-shot=true, training-free=true, features=diffusion2026.01 | 44.2 | — | |
| MaskCutzero-shot=true, training-free=true, features=diffusion2026.01 | 41.7 | — | |
| EQUSSEvaluation Protocol=Linear Probe, Backbone=ViT-S2023.12 | 41.2 | 75.2 | |
| STROP-SEncoder=DINO-B, Representation=latent, Evaluation protocol=Linear probing, Training dataset=COCO2026.05 | 39.2 | — | |
| STEGOEvaluation Protocol=Linear Probe, Backbone=ViT-S2023.12 | 38.3 | 74.4 | |
| DINO-SEncoder=DINO-S, Model=Teacher (frozen DINO), Repr.=teacher, Evaluation Protocol=Linear probe2026.05 | 36.3 | — | |
| DINO-SRepresentation=teacher, Evaluation protocol=Linear probing2026.05 | 36.3 | — | |
| STROP-MEncoder=DINO-B, Representation=latent, Evaluation protocol=Linear probing, Training dataset=COCO2026.05 | 36.3 | — | |
| STROP-MEncoder=DINO-B, Model=STROP-M, Repr.=latent, Evaluation Protocol=Linear probe2026.05 | 35.5 | — | |
| STROP-MEncoder=DINO-L, Model=STROP-M, Repr.=latent, Evaluation Protocol=Linear probe2026.05 | 35.1 | — | |
| STROP-SEncoder=DINO-S, Representation=latent, Evaluation protocol=Linear probing, Training dataset=COCO2026.05 | 32.5 | — | |
| STROP-SEncoder=DINO-B, Model=STROP-S, Repr.=latent, Evaluation Protocol=Linear probe2026.05 | 32.2 | — | |
| STROP-SEncoder=DINO-S, Model=STROP-S, Repr.=latent, Evaluation Protocol=Linear probe2026.05 | 29.8 | — | |
| DINOEvaluation Protocol=Linear Probe2023.12 | 29.4 | 66.8 | |
| STROP-MEncoder=DINO-S, Model=STROP-M, Repr.=latent, Evaluation Protocol=Linear probe2026.05 | 28.8 | — | |
| STEGOLearning type=Unsupervised2023.02 | 26.8 | — | |
| ReCOzero-shot=true, training-free=true, features=diffusion2026.01 | 26.3 | — | |
| EQUSSEvaluation Protocol=Unsupervised, Backbone=ViT-S2023.12 | 25.8 | 53.8 | |
| STEGOEvaluation Protocol=Unsupervised, Backbone=ViT-S2023.12 | 24.5 | 48.3 | |
| Cyclic walksLearning type=Unsupervised2023.02 | 22.5 | — | |
| FlexTokEncoder=FlexTok, Model=D12-D12-IN1k, Repr.=latent, Evaluation Protocol=Linear probe2026.05 | 21.6 | — | |
| FlexTokModel=D12-D12-IN1k, Representation=latent, Evaluation protocol=Linear probing2026.05 | 21.6 | — | |
| STROP-SEncoder=DINO-S, Representation=program, Evaluation protocol=Linear probing, Training dataset=COCO2026.05 | 20 | — | |
| MaskCLIPzero-shot=true, training-free=true, features=diffusion2026.01 | 19.6 | — | |
| STROP-MEncoder=DINO-B, Model=STROP-M, Repr.=program, Evaluation Protocol=Linear probe2026.05 | 19.6 | — | |
| STROP-SEncoder=DINO-B, Representation=program, Evaluation protocol=Linear probing, Training dataset=COCO2026.05 | 19.5 | — | |
| STROP-SEncoder=DINO-B, Model=STROP-S, Repr.=program, Evaluation Protocol=Linear probe2026.05 | 18.9 | — | |
| STROP-SEncoder=DINO-S, Model=STROP-S, Repr.=program, Evaluation Protocol=Linear probe2026.05 | 18.4 | — | |
| SlotConLearning type=Unsupervised2023.02 | 18.3 | — | |
| STROP-MEncoder=DINO-B, Representation=program, Evaluation protocol=Linear probing, Training dataset=COCO2026.05 | 18 | — | |
| STROP-MEncoder=DINO-L, Model=STROP-M, Repr.=program, Evaluation Protocol=Linear probe2026.05 | 17.9 | — | |
| TransFGUEvaluation Protocol=Unsupervised, Backbone=ViT-S2023.12 | 17.5 | 52.7 | |
| STROP-MEncoder=DINO-S, Model=STROP-M, Repr.=program, Evaluation Protocol=Linear probe2026.05 | 17.2 | — | |
| ACSegLearning type=Unsupervised2023.02 | 16.4 | — | |
| ACSegEvaluation Protocol=Unsupervised2023.12 | 16.4 | — | |
| FlexTokEncoder=FlexTok, Model=D12-D12-IN1k, Repr.=program, Evaluation Protocol=Linear probe2026.05 | 15.6 | — | |
| PiCIE+HEvaluation Protocol=Linear Probe2023.12 | 14.8 | 54.8 | |
| PiCIE+H2022.10 | 14.4 | — | |
| PiCIE + HLearning type=Unsupervised2023.02 | 14.4 | — | |
| PiCIE+HEvaluation Protocol=Unsupervised2023.12 | 14.4 | 50 | |
| PiCIEEvaluation Protocol=Linear Probe2023.12 | 13.9 | 54.2 | |
| PiCIE2022.10 | 13.8 | — | |
| PiCIEEvaluation Protocol=Unsupervised2023.12 | 13.8 | 48.1 | |
| MDCEvaluation Protocol=Linear Probe2023.12 | 13.3 | 48.6 | |
| MoCoV2Evaluation Protocol=Linear Probe2023.12 | 13.2 | 44.4 | |
| One-D-Piece-S-256Encoder=DINOv3-S, Model=One-D-Piece-S-256, Repr.=latent, Evaluation Protocol=Linear probe2026.05 | 10.6 | — | |
| One-D-PieceEncoder=DINOv3-S, Model=One-D-Piece-S-256, Representation=latent, Evaluation protocol=Linear probing2026.05 | 10.6 | — | |
| MoCoV2Evaluation Protocol=Unsupervised2023.12 | 10.4 | 25.2 | |
| ResNet50Evaluation Protocol=Linear Probe, Backbone=ResNet-502023.12 | 10.2 | 41.3 | |
| Modified DC2022.10 | 9.8 | — | |
| MDCEvaluation Protocol=Unsupervised2023.12 | 9.8 | 32.2 | |
| DINO2022.10 | 9.6 | — | |
| DINOEvaluation Protocol=Unsupervised2023.12 | 9.6 | 30.5 | |
| ImageNet2022.10 | 8.9 | — | |
| ResNet50Evaluation Protocol=Unsupervised, Backbone=ResNet-502023.12 | 8.9 | 24.6 | |
| IICEvaluation Protocol=Linear Probe2023.12 | 8.4 | 44.5 | |
| One-D-Piece-S-256Encoder=DINOv3-S, Model=One-D-Piece-S-256, Repr.=program, Evaluation Protocol=Linear probe2026.05 | 7.2 | — | |
| IIC2022.10 | 6.7 | — | |
| IICEvaluation Protocol=Unsupervised2023.12 | 6.7 | 21.8 | |
| MoCo v22022.10 | 4.4 | — | |
| ACEvaluation Protocol=Unsupervised2023.12 | — | 30.8 | |
| Deep ClusterEvaluation Protocol=Unsupervised2023.12 | — | 19.9 | |
| Isola et al.Evaluation Protocol=Unsupervised2023.12 | — | 24.3 | |
| SIFTEvaluation Protocol=Unsupervised2023.12 | — | 20.2 |