Image Classification on Country211 (Accuracy, ∆CA, ASR)
31.8AccuracyCLIP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CLIPBackbone=ViT-L/14, Zero-shot=true2025.07 | 31.8 | — | — | |
| LongCLIPBackbone=ViT-L/14, Zero-shot=true2025.07 | 28.1 | — | — | |
| SmartCLIPBackbone=ViT-L/14, Zero-shot=true2025.07 | 26.9 | — | — | |
| AXISArchitecture=ViT-B-322025.08 | 18.28 | — | — | |
| aTLASArchitecture=ViT-B-322025.08 | 18.08 | — | — | |
| VLM BaselineIR=1, Backbone=VLM2026.01 | 17.2 | — | — | |
| CLIP2026.04 | 15.26 | — | — | |
| FARE2026.04 | 12.46 | — | — | |
| AdvFLYPfull2026.04 | 11 | — | — | |
| AdvFLYP2026.04 | 10.54 | — | — | |
| TGA-ZSR2026.04 | 9.86 | — | — | |
| PMG-AFT2026.04 | 9.23 | — | — | |
| TeCoA2026.04 | 8.14 | — | — | |
| RN50 BaselineIR=1, Backbone=ResNet-502026.01 | 6.8 | — | — | |
| HyCoCLIPBackbone=ViT-B/16, Zero-shot=true2026.03 | 6.5 | — | — | |
| UNCHABackbone=ViT-B/16, Zero-shot=true2026.03 | 6.1 | — | — | |
| CLIPBackbone=ViT-B/16, Zero-shot=true2026.03 | 5.7 | — | — | |
| MERUBackbone=ViT-B/16, Zero-shot=true2026.03 | 5.4 | — | — | |
| MERUBackbone=ViT-S/16, Zero-shot=true2026.03 | 5.3 | — | — | |
| HyCoCLIPBackbone=ViT-S/16, Zero-shot=true2026.03 | 5.2 | — | — | |
| UNCHABackbone=ViT-S/16, Zero-shot=true2026.03 | 5.2 | — | — | |
| CLIPBackbone=ViT-S/16, Zero-shot=true2026.03 | 5.1 | — | — | |
| ATMG†Backbone=ViT-S/16, Zero-shot=true2026.03 | 4.6 | — | — | |
| ATMG†Backbone=ViT-B/16, Zero-shot=true2026.03 | 4.6 | — | — | |
| PRISMIR=1, Attack Type=Blend, Backbone=ResNet-502026.01 | — | 5.3 | 0 | |
| PRISMIR=1, Attack Type=BPP, Backbone=ResNet-502026.01 | — | 5.4 | 1.4 | |
| PRISMIR=1, Attack Type=CTRL, Backbone=ResNet-502026.01 | — | 5.3 | 0 |