Multi-label Classification on VOC 07
97mAPML-VPT
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ML-VPTBackbone=DINOv2/B, Resolution=448 x 4482025.04 | 97 | 92.1 | 93.1 | — | — | — | |
| ML-VPTBackbone=DINOv2/B, Resolution=224 x 2242025.04 | 96.4 | 91 | 92.3 | — | — | — | |
| Gate VPTBackbone=DINOv2/B, Resolution=448 x 4482025.04 | 96.2 | 91.1 | 92.5 | — | — | — | |
| VPTBackbone=DINOv2/B, Resolution=448 x 4482025.04 | 96.1 | 91 | 92.3 | — | — | — | |
| E2VPTBackbone=DINOv2/B, Resolution=448 x 4482025.04 | 96.1 | 91.1 | 92.4 | — | — | — | |
| ML-VPTBackbone=DINOv2/S, Resolution=448 x 4482025.04 | 95.7 | 90.2 | 92 | — | — | — | |
| ML-VPTBackbone=ViT-B-21k, Resolution=448 x 4482025.04 | 95.6 | — | 91.3 | — | — | — | |
| VPTBackbone=DINOv2/B, Resolution=224 x 2242025.04 | 95.6 | 90.2 | 91.6 | — | — | — | |
| Gate VPTBackbone=DINOv2/B, Resolution=224 x 2242025.04 | 95.5 | 89.7 | 91.2 | — | — | — | |
| E2VPTBackbone=DINOv2/B, Resolution=224 x 2242025.04 | 95.4 | 89.9 | 91.4 | — | — | — | |
| Gate VPTBackbone=ViT-B-21k, Resolution=448 x 4482025.04 | 95.2 | 89.6 | 91.1 | — | — | — | |
| E2VPTBackbone=ViT-B-21k, Resolution=448 x 4482025.04 | 95.1 | 89.3 | 90.7 | — | — | — | |
| ML-VPTBackbone=ViT-B-21k, Resolution=224 x 2242025.04 | 95 | 89.2 | 90.3 | — | — | — | |
| VPTBackbone=ViT-B-21k, Resolution=448 x 4482025.04 | 94.4 | 88.7 | 90.4 | — | — | — | |
| ML-VPTBackbone=ViT-B, Resolution=448 x 4482025.04 | 94.3 | 88.6 | 90.1 | — | — | — | |
| Gate VPTBackbone=ViT-B-21k, Resolution=224 x 2242025.04 | 94.3 | 88.5 | 89.7 | — | — | — | |
| ML-VPTBackbone=DINOv2/S, Resolution=224 x 2242025.04 | 94.3 | 88.3 | 90.3 | — | — | — | |
| VPTBackbone=ViT-B-21k, Resolution=224 x 2242025.04 | 94.2 | 88 | 89.6 | — | — | — | |
| E2VPTBackbone=ViT-B-21k, Resolution=224 x 2242025.04 | 94.1 | 87.9 | 89.4 | — | — | — | |
| VPTBackbone=DINOv2/S, Resolution=448 x 4482025.04 | 93.8 | 88.1 | 90.1 | — | — | — | |
| SCPNetPartial label proportion=Avg.2023.03 | 93.5 | — | — | — | — | — | |
| VPTBackbone=ViT-B, Resolution=448 x 4482025.04 | 93.4 | 87.4 | 89.1 | — | — | — | |
| Gate VPTBackbone=DINOv2/S, Resolution=448 x 4482025.04 | 93.3 | 87.3 | 89.3 | — | — | — | |
| DualCoOpPartial label proportion=Avg.2023.03 | 93.2 | — | — | — | — | — | |
| ML-VPTBackbone=ViT-B, Resolution=224 x 2242025.04 | 92.9 | 86.4 | 88.1 | — | — | — | |
| VPTBackbone=DINOv2/S, Resolution=224 x 2242025.04 | 92.8 | 86.4 | 88.5 | — | — | — | |
| ML-VPTBackbone=MoCo v3, Resolution=448 x 4482025.04 | 92.7 | 86.3 | 88.6 | — | — | — | |
| E2VPTBackbone=ViT-B, Resolution=448 x 4482025.04 | 92.4 | 86.5 | 88.3 | — | — | — | |
| Gate VPTBackbone=DINOv2/S, Resolution=224 x 2242025.04 | 92.1 | 85.5 | 87.7 | — | — | — | |
| E2VPTBackbone=DINOv2/S, Resolution=448 x 4482025.04 | 91.2 | 84.8 | 88 | — | — | — | |
| Gate VPTBackbone=ViT-B, Resolution=448 x 4482025.04 | 91.1 | 85.2 | 87.3 | — | — | — | |
| ML-VPTBackbone=MoCo v3, Resolution=224 x 2242025.04 | 91.1 | 84.2 | 86.8 | — | — | — | |
| SARBPartial label proportion=Avg.2023.03 | 90.7 | — | — | — | — | — | |
| VPTBackbone=ViT-B, Resolution=224 x 2242025.04 | 90.4 | 83.4 | 85.7 | — | — | — | |
| ML-VPTBackbone=MAE, Resolution=448 x 4482025.04 | 90.4 | 83.9 | 87.4 | — | — | — | |
| SSTPartial label proportion=Avg.2023.03 | 90.4 | — | — | — | — | — | |
| E2VPTBackbone=DINOv2/S, Resolution=224 x 2242025.04 | 89.6 | 83.1 | 86.3 | — | — | — | |
| SSGRLPartial label proportion=Avg.2023.03 | 89.5 | — | — | — | — | — | |
| FedMPTMask=10%, Venue=Ours2026.05 | 89.4 | 83.06 | 81.12 | — | — | — | |
| ML-VPTBackbone=MAE, Resolution=224 x 2242025.04 | 89.2 | 82.7 | 86.4 | — | — | — | |
| SwAVEvaluation protocol=linear classifier on fixed features, multi-crop=true2021.03 | 88.9 | — | — | — | — | — | |
| GCN-MLPartial label proportion=Avg.2023.03 | 88.9 | — | — | — | — | — | |
| E2VPTBackbone=MoCo v3, Resolution=448 x 4482025.04 | 88.8 | 82.4 | 85.6 | — | — | — | |
| VPTBackbone=MoCo v3, Resolution=448 x 4482025.04 | 88.4 | 82.3 | 85.7 | — | — | — | |
| Fed-RAMMask=10%, Venue=CVPR’252026.05 | 87.89 | 77.03 | 75.93 | — | — | — | |
| Gate VPTBackbone=MoCo v3, Resolution=448 x 4482025.04 | 87.7 | 81.1 | 84.7 | — | — | — | |
| SupervisedEvaluation protocol=linear classifier on fixed features2021.03 | 87.5 | — | — | — | — | — | |
| Gate VPTBackbone=ViT-B, Resolution=224 x 2242025.04 | 87.4 | 80.9 | 82.9 | — | — | — | |
| BYOLEvaluation protocol=linear classifier on fixed features2021.03 | 86.6 | — | — | — | — | — | |
| MoCo-v2Evaluation protocol=linear classifier on fixed features2021.03 | 86.4 | — | — | — | — | — | |
| SwAVEvaluation protocol=linear classifier on fixed features, multi-crop=false2021.03 | 86.4 | — | — | — | — | — | |
| VPTBackbone=MoCo v3, Resolution=224 x 2242025.04 | 86.2 | 79.7 | 83.6 | — | — | — | |
| BARLOW TWINSEvaluation protocol=linear classifier on fixed features2021.03 | 86.2 | — | — | — | — | — | |
| Gate VPTBackbone=MAE, Resolution=448 x 4482025.04 | 86.1 | 79.2 | 84.2 | — | — | — | |
| E2VPTBackbone=MoCo v3, Resolution=224 x 2242025.04 | 85.9 | 79.2 | 83.1 | — | — | — | |
| E2VPTBackbone=ViT-B, Resolution=224 x 2242025.04 | 85.8 | 79.4 | 83.7 | — | — | — | |
| SimCLREvaluation protocol=linear classifier on fixed features2021.03 | 85.5 | — | — | — | — | — | |
| FedMVPMask=10%, Venue=ICCV’252026.05 | 85.27 | 79.1 | 76.23 | — | — | — | |
| FedMPTMask=Avg, Venue=Ours2026.05 | 85.19 | 78.72 | 76.34 | — | — | — | |
| Gate VPTBackbone=MAE, Resolution=224 x 2242025.04 | 85.1 | 75.4 | 80.5 | — | — | — | |
| E2VPTBackbone=MAE, Resolution=448 x 4482025.04 | 84.6 | 78.2 | 83.2 | — | — | — | |
| FedTPGMask=10%, Venue=ICLR’242026.05 | 84.19 | 77.45 | 76.93 | — | — | — | |
| Gate VPTBackbone=MoCo v3, Resolution=224 x 2242025.04 | 83.5 | 77.1 | 81.2 | — | — | — | |
| E2VPTBackbone=MAE, Resolution=224 x 2242025.04 | 83.3 | 76.7 | 81.5 | — | — | — | |
| VPTBackbone=MAE, Resolution=448 x 4482025.04 | 82.8 | 75.9 | 81.6 | — | — | — | |
| Fed-DualCoOpMask=10%, Venue=NeurIPS’222026.05 | 82.52 | 76.86 | 76.58 | — | — | — | |
| Fed-MaPLeMask=10%, Venue=CVPR’232026.05 | 81.27 | 75.59 | 73.49 | — | — | — | |
| Fed-PosCoOpMask=10%, Venue=WACV’252026.05 | 81.18 | 76.44 | 73.29 | — | — | — | |
| FedAWAMask=10%, Venue=CVPR’252026.05 | 80.81 | 70.7 | 72.94 | — | — | — | |
| VPTBackbone=MAE, Resolution=224 x 2242025.04 | 80.3 | 74.1 | 79.2 | — | — | — | |
| FedPGPMask=10%, Venue=ICML’242026.05 | 79.96 | 73.81 | 63.81 | — | — | — | |
| Fed-SCPNetMask=10%, Venue=CVPR’232026.05 | 78.81 | 70.83 | 72.6 | — | — | — | |
| Fed-TCPMask=10%, Venue=CVPR’242026.05 | 76.42 | 76 | 72.5 | — | — | — | |
| ML-GCNCondition=w.2025.06 | — | — | — | 92.1 | 50.5 | 92.3 | |
| ML-GCNCondition=w/o.2025.06 | — | — | — | 90.2 | 81.1 | 93.4 | |
| ML-GCNCondition=w. ∧ w/o.2025.06 | — | — | — | 88.7 | 46.4 | 90.4 | |
| MLDecoderCondition=w.2025.06 | — | — | — | 94.2 | 32.2 | 97.6 | |
| MLDecoderCondition=w/o.2025.06 | — | — | — | 93.8 | 80.1 | 87.1 | |
| MLDecoderCondition=w. ∧ w/o.2025.06 | — | — | — | 93.8 | 31 | 84.9 |