Multi-label Image Classification on MS-COCO
90.3mAPQ2L-TResL
Evaluation Results
| Method | Links | |
|---|---|---|
| Q2L-TResLArchitecture=TResNetL(22k), Input Resolution=640x6402021.07 | 90.3 | |
| TResNetArchitecture=TResNetL(22k), Input Resolution=640x6402021.07 | 89.8 | |
| Q2L-TResXLArchitecture=TResNetXL, Input Resolution=640x6402021.07 | 89 | |
| ASLArchitecture=TResNetXL, Input Resolution=640x6402021.07 | 88.4 | |
| IN1k-Mul E2EBackbone=ViT-B/16, Resolution=224, Supervision=Multi-label (Mul), Training Protocol=End-to-end2026.03 | 84.7 | |
| IN21k-Mul PretrainBackbone=ViT-B/16, Resolution=224, Pre-training=ImageNet-21K (MIIL), Supervision=Multi-label (Mul)2026.03 | 82.8 | |
| D2ACEArchitecture=HST2026.05 | 82.52 | |
| +IN1k-Mul FTBackbone=ViT-B/16, Resolution=224, Pre-training=ImageNet-21K (MIIL), Supervision=Multi-label (Mul), Training Protocol=Fine-tuning2026.03 | 82.5 | |
| +IN1k-Sig FTBackbone=ViT-B/16, Resolution=224, Pre-training=ImageNet-21K (MIIL), Supervision=Single-label (Sig), Training Protocol=Fine-tuning2026.03 | 82.3 | |
| DIHCLArchitecture=HST2026.05 | 81.86 | |
| Hard-ImbArchitecture=HST2026.05 | 81.82 | |
| ML-UncArchitecture=HST2026.05 | 81.79 | |
| D2ACEArchitecture=ML-GCN2026.05 | 81.42 | |
| ImageNet-21K PretrainPre-training Dataset=ImageNet-21K, Pre-training Method=Semantic Softmax2021.04 | 81.3 | |
| RecentArchitecture=HST2026.05 | 81.27 | |
| DIHCLArchitecture=ML-GCN2026.05 | 81.03 | |
| ML-UncArchitecture=ML-GCN2026.05 | 80.85 | |
| RecentArchitecture=ML-GCN2026.05 | 80.74 | |
| BalanceArchitecture=HST2026.05 | 80.65 | |
| Hard-ImbArchitecture=ML-GCN2026.05 | 80.55 | |
| ActiveArchitecture=HST2026.05 | 80.54 | |
| Open Images PretrainPre-training Dataset=Open Images (v6), Pre-training Method=Multi-label training2021.04 | 80.5 | |
| ActiveArchitecture=ML-GCN2026.05 | 80.25 | |
| BalanceArchitecture=ML-GCN2026.05 | 80.21 | |
| RandomArchitecture=HST2026.05 | 80.12 | |
| RandomArchitecture=ML-GCN2026.05 | 79.62 |