Image Classification on DTD (test) (Accuracy)
69.42Accuracy (DTD Test)AP
Evaluation Results
| Method | Links | |
|---|---|---|
| APArchitecture=CLIP (ViT-B/32)2026.04 | 69.42 | |
| APArchitecture=Swin-Transformer (Swin-B)2026.04 | 69.13 | |
| VPArchitecture=Swin-Transformer (Swin-B)2026.04 | 65.39 | |
| VPArchitecture=CLIP (ViT-B/32)2026.04 | 64.43 | |
| UR-JEPA(L CGLT)Pre-train dataset=Inet100, Pre-train epochs=800, Evaluation protocol=frozen-backbone linear-probe, Backbone features=concat of the last two CLS tokens, Probe epochs=50, Probe optimizer=AdamW with cosine annealing2026.05 | 59.41 | |
| LeJEPA(L SIGReg)Pre-train dataset=Inet100, Pre-train epochs=800, Evaluation protocol=frozen-backbone linear-probe, Backbone features=concat of the last two CLS tokens, Probe epochs=50, Probe optimizer=AdamW with cosine annealing2026.05 | 58.67 | |
| DMN-ZSBackbone=ViT-B/162026.06 | 55.85 | |
| ComMemBackbone=ViT-B/162026.06 | 55.5 | |
| DPEBackbone=ViT-B/162026.06 | 54.2 | |
| ComMemBackbone=ResNet-502026.06 | 53.43 | |
| HopSBackbone (Insd)=ResNet-50, Insd confusion rate (γc)=0.67, Prompting (uni-prompt)=true2026.04 | 51.7 | |
| TPSBackbone=ViT-B/162026.06 | 50.47 | |
| DMN-ZSBackbone=ResNet-502026.06 | 50.41 | |
| DPEBackbone=ResNet-502026.06 | 50.18 | |
| HopSBackbone (Insd)=ResNet-18, Insd confusion rate (γc)=0.67, Prompting (uni-prompt)=true2026.04 | 48.8 | |
| DynaPromptBackbone=ViT-B/162026.06 | 47.96 | |
| HopSBackbone (Insd)=CLIP-ResNet50, Insd confusion rate (γc)=0.67, Prompting (uni-prompt)=true2026.04 | 47.8 | |
| TPTBackbone=ViT-B/162026.06 | 47.75 | |
| TDABackbone=ViT-B/162026.06 | 47.4 | |
| HopSBackbone (Insd)=ResNet-18, Insd confusion rate (γc)=0.75, Prompting (uni-prompt)=true2026.04 | 47.3 | |
| DiffTPTBackbone=ViT-B/162026.06 | 47 | |
| HopSBackbone (Insd)=ResNet-50, Insd confusion rate (γc)=0.75, Prompting (uni-prompt)=true2026.04 | 46.2 | |
| EnsembleBackbone=ViT-B/162026.06 | 45.04 | |
| CLIPBackbone=ViT-B/162026.06 | 44.27 | |
| TDABackbone=ResNet-502026.06 | 43.74 | |
| HopSBackbone=ResNet-18, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 43.1 | |
| HopSBackbone=ResNet-50, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 42.5 | |
| CoOpBackbone=ViT-B/162026.06 | 41.92 | |
| HopSBackbone (Insd)=CLIP-ResNet50, Insd confusion rate (γc)=0.75, Prompting (uni-prompt)=true2026.04 | 41.4 | |
| DynaPromptBackbone=ResNet-502026.06 | 41.19 | |
| TPTBackbone=ResNet-502026.06 | 40.84 | |
| DiffTPTBackbone=ResNet-502026.06 | 40.72 | |
| CLIPBackbone=ResNet-502026.06 | 40.37 | |
| EnsembleBackbone=ResNet-502026.06 | 40.37 | |
| DSPTNOISE TYPE=SYM 90%2026.05 | 40.02 | |
| LWCBackbone=CLIP-ResNet50, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 38.1 | |
| LWCBackbone=ResNet-18, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 37.8 | |
| HopSBackbone=CLIP-ResNet50, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 37.4 | |
| CoOpBackbone=ResNet-502026.06 | 37.29 | |
| NLPROMTNOISE TYPE=SYM 90%2026.05 | 36.6 | |
| LWCBackbone=ResNet-50, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 34.5 | |
| DSPTNOISE TYPE=PAIR 80%2026.05 | 33.66 | |
| CCBackbone=ResNet-18, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 32.2 | |
| CEBackbone=ResNet-50, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 32 | |
| CEBackbone=ResNet-18, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 31.2 | |
| NLPROMTNOISE TYPE=PAIR 80%2026.05 | 30.93 | |
| CEBackbone=CLIP-ResNet50, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 30.9 | |
| SELECTNOISE TYPE=SYM 90%2026.05 | 29.91 | |
| LOGITNORMNOISE TYPE=SYM 90%2026.05 | 29.08 | |
| CCBackbone=ResNet-50, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 25.8 | |
| CCBackbone=CLIP-ResNet50, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 25.7 | |
| BOOTSTRAPNOISE TYPE=SYM 90%2026.05 | 22.36 | |
| COOPNOISE TYPE=SYM 90%2026.05 | 21.19 | |
| SMOOTHINGNOISE TYPE=SYM 90%2026.05 | 19.75 | |
| EXPBackbone=CLIP-ResNet50, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 19.4 | |
| EXPBackbone=ResNet-18, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 19.1 | |
| EXPBackbone=ResNet-50, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 16.7 | |
| GCEBackbone=ResNet-18, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 16.6 | |
| GCEBackbone=ResNet-50, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 15.1 | |
| COOPNOISE TYPE=PAIR 80%2026.05 | 8.58 | |
| GCEBackbone=CLIP-ResNet50, Confusion rate=0.80, Prompt type=uni-prompt2026.04 | 7.6 | |
| BOOTSTRAPNOISE TYPE=PAIR 80%2026.05 | 6.99 | |
| SMOOTHINGNOISE TYPE=PAIR 80%2026.05 | 5.78 | |
| SELECTNOISE TYPE=PAIR 80%2026.05 | 5.44 | |
| LOGITNORMNOISE TYPE=PAIR 80%2026.05 | 1.48 |