Semantic Segmentation on Cityscapes (1/8 subset, val split)
84.3mIoUUniMatch V2
Evaluation results
| Method | Links | |
|---|---|---|
| UniMatch V2Venue=Ours, Encoder=DINOV2-B, #Params=97.5M2024.10 | 84.3 | |
| UniMatchV2-B + PrevMatchBackbone=DINOv2-B, Resolution=756x7562024.05 | 84.1 | |
| UniMatchV2-BBackbone=DINOv2-B, Resolution=756x7562024.05 | 83.3 | |
| UniMatchV2-S + PrevMatchBackbone=DINOv2-S2024.05 | 82.7 | |
| Labeled OnlyEncoder=DINOV2-B, #Params=97.5M2024.10 | 82.7 | |
| UniMatchV2-SBackbone=DINOv2-S2024.05 | 81.9 | |
| UniMatch V2Venue=Ours, Encoder=DINOV2-S, #Params=24.8M2024.10 | 81.9 | |
| Labeled OnlyEncoder=DINOV2-S, #Params=24.8M2024.10 | 80.2 | |
| SemiVLBackbone=Transformer-based2024.05 | 79.4 | |
| SemiVLVenue=ECCV'24, Encoder=CLIP-B, #Params=88.0M2024.10 | 79.4 | |
| AllSparkBackbone=Transformer-based2024.05 | 79.2 | |
| AllSparkVenue=CVPR'24, Encoder=MiT-B5, #Params=89.3M2024.10 | 79.2 | |
| BeyondPixelsVenue=ECCV'24, Encoder=RN-101, #Params=59.5M2024.10 | 79.2 | |
| AugSeg + PrevMatchBackbone=ResNet-1012024.05 | 78.9 | |
| UniMatch + PrevMatchBackbone=ResNet-1012024.05 | 78.9 | |
| LogicDiagVenue=ICCV'23, Encoder=RN-101, #Params=59.5M2024.10 | 78.9 | |
| CorrMatchBackbone=ResNet-1012024.05 | 78.5 | |
| CorrMatchVenue=CVPR'24, Encoder=RN-101, #Params=59.5M2024.10 | 78.5 | |
| Dual TeacherBackbone=ResNet-1012024.05 | 78.4 | |
| DAWVenue=NeurIPS'23, Encoder=RN-101, #Params=59.5M2024.10 | 78.4 | |
| UniMatch V1Venue=CVPR'23, Encoder=CLIP-B, #Params=88.0M2024.10 | 78.2 | |
| DDFPVenue=CVPR'24, Encoder=RN-101, #Params=59.5M2024.10 | 78.2 | |
| UniMatchBackbone=ResNet-1012024.05 | 77.9 | |
| UniMatch V1Venue=CVPR'23, Encoder=RN-101, #Params=59.5M2024.10 | 77.9 | |
| AugSegBackbone=ResNet-1012024.05 | 77.8 | |
| AugSegVenue=CVPR'23, Encoder=RN-101, #Params=59.5M2024.10 | 77.8 | |
| Diverse Co-T.Backbone=ResNet-1012024.05 | 77.4 | |
| iMASVenue=CVPR'23, Encoder=RN-101, #Params=59.5M2024.10 | 77.4 | |
| Diverse CoTVenue=ICCV'23, Encoder=RN-101, #Params=59.5M2024.10 | 77.4 | |
| ESLVenue=ICCV'23, Encoder=RN-101, #Params=59.5M2024.10 | 77.2 | |
| PS-MTBackbone=ResNet-1012024.05 | 76.9 | |
| PS-MTVenue=CVPR'22, Encoder=RN-101, #Params=59.5M2024.10 | 76.9 | |
| U2PLVenue=CVPR'22, Encoder=RN-101, #Params=59.5M2024.10 | 76.5 | |
| CCVCBackbone=ResNet-50, Crop Size=712, Partition Protocol=1/82023.03 | 76.4 | |
| PCRVenue=NeurIPS'22, Encoder=RN-101, #Params=59.5M2024.10 | 76.3 | |
| PS-MTBackbone=ResNet-50, Crop Size=712, Partition Protocol=1/82023.03 | 75.8 | |
| USRNBackbone=ResNet-50, Crop Size=712, Partition Protocol=1/82023.03 | 75 | |
| CPSBackbone=ResNet-50, Crop Size=712, Partition Protocol=1/8, reproduced by=U2PL2023.03 | 74.3 | |
| U2PLBackbone=ResNet-50, Crop Size=712, Partition Protocol=1/82023.03 | 73 | |
| Labeled OnlyEncoder=RN-101, #Params=59.5M2024.10 | 72.8 | |
| ST++Architecture=DeepLabv3+, Backbone=ResNet-502021.06 | 72.7 | |
| ST++Backbone=ResNet-50, Crop Size=712, Partition Protocol=1/82023.03 | 72.7 | |
| CCTBackbone=ResNet-50, Crop Size=712, Partition Protocol=1/82023.03 | 72.5 | |
| GTA-SegVenue=NeurIPS'22, Encoder=RN-101, #Params=59.5M2024.10 | 72 | |
| STArchitecture=DeepLabv3+, Backbone=ResNet-502021.06 | 71.6 | |
| GCTBackbone=ResNet-50, Crop Size=712, Partition Protocol=1/82023.03 | 71.3 | |
| ELNBackbone=ResNet-50, Crop Size=712, Partition Protocol=1/82023.03 | 70.3 | |
| PseudoSegArchitecture=DeepLabv3+, Backbone=ResNet-1012021.06 | 69.8 | |
| DCCArchitecture=DeepLabv3+, Backbone=ResNet-502021.06 | 69.7 | |
| GuidedMix-Net2021.06 | 65.8 | |
| CutMix-SegArchitecture=DeepLabv3+, Backbone=ResNet-1012021.06 | 65.8 | |
| SupOnlyArchitecture=DeepLabv3+, Backbone=ResNet-502021.06 | 65.8 | |
| Supervised BaselineBackbone=ResNet-50, Crop Size=712, Partition Protocol=1/82023.03 | 65.8 | |
| DMTArchitecture=DeepLabv3+, Backbone=ResNet-1012021.06 | 63 | |
| ClassMix2021.06 | 61.4 | |
| ClassMixArchitecture=DeepLabv3+, Backbone=ResNet-1012021.06 | 61.4 | |
| CutMixBackbone=DeepLab v2, Pre-training=ImageNet, Labeled samples=1/8 (372), Number of runs=52019.06 | 60.34 | |
| CutMix2021.06 | 60.3 | |
| s4GANBackbone=DeepLab v2, Pre-training=ImageNet, Labeled samples=1/8 (372)2019.06 | 59.3 | |
| s4GAN2021.06 | 59.3 | |
| CutoutBackbone=DeepLab v2, Pre-training=ImageNet, Labeled samples=1/8 (372), Number of runs=52019.06 | 57.72 | |
| AdversarialBackbone=DeepLab v2, Pre-training=ImageNet, Labeled samples=1/8 (372)2019.06 | 57.1 | |
| AdvSSL2021.06 | 57.1 | |
| Baseline (Mittal et al.)Backbone=DeepLab v2, Pre-training=ImageNet, Labeled samples=1/8 (372)2019.06 | 56.2 | |
| Baseline (Ours)Backbone=DeepLab v2, Pre-training=ImageNet, Labeled samples=1/8 (372), Number of runs=52019.06 | 55.25 |