Attribution Evaluation on ImageNet (val)
18.1POS ScoreDEX-AR
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| DEX-ARBase Model=BakLlaVA-7B, Inference Time (sec./img.)=0.712026.03 | 18.1 | 2.48 | 72 | 13 | 50 | — | — | — | |
| Integrated Grad.Base Model=BakLlaVA-7B, Inference Time (sec./img.)=3.252026.03 | 13.5 | 12.77 | 44 | 32 | 33 | — | — | — | |
| Attn x CAMBase Model=BakLlaVA-7B, Inference Time (sec./img.)=0.652026.03 | 12.6 | 3.74 | 65 | 18 | 42 | — | — | — | |
| CheferCAMBase Model=BakLlaVA-7B, Inference Time (sec./img.)=6.202026.03 | 11.15 | 4.07 | 64 | 19 | 36 | — | — | — | |
| IIABase Model=BakLlaVA-7B, Inference Time (sec./img.)=15.302026.03 | 11.08 | 3.77 | 65 | 18 | 40 | — | — | — | |
| GradCAMBase Model=BakLlaVA-7B, Inference Time (sec./img.)=0.662026.03 | 7.49 | 4.39 | 64 | 20 | 28 | — | — | — | |
| RISEBase Model=BakLlaVA-7B, Inference Time (sec./img.)=11.802026.03 | 6.39 | 3.87 | 65 | 21 | 28 | — | — | — | |
| RolloutBase Model=BakLlaVA-7B, Inference Time (sec./img.)=0.512026.03 | 6.13 | 3.12 | 69 | 14 | 21 | — | — | — | |
| Grad-CAMClassifier=ViT-B/32, Attribution Target=Predicted Class, Runtime=0.010s2026.04 | 0.293 | 0.6328 | 8.5072 | 52.94 | 23.37 | 29.56 | 0.3398 | 66.9314 | |
| Trans-Att.Classifier=ViT-B/32, Attribution Target=Predicted Class, Runtime=0.157s2026.04 | 0.2567 | 0.6534 | 34.978 | 55.07 | 20.24 | 34.83 | 0.3967 | 17.1553 | |
| ViT-CXClassifier=ViT-B/32, Attribution Target=Predicted Class, Runtime=0.763s2026.04 | 0.2525 | 0.6622 | 48.995 | 56.04 | 19.82 | 36.22 | 0.4098 | 9.3272 | |
| MDAClassifier=ViT-B/32, Attribution Target=Predicted Class, Runtime=1.803s2026.04 | 0.2511 | 0.7043 | 26.1375 | 59.39 | 19.73 | 39.66 | 0.4533 | 25.2806 | |
| LeGradClassifier=ViT-B/32, Attribution Target=Predicted Class, Runtime=0.005s2026.04 | 0.2457 | 0.6521 | 39.4919 | 55.02 | 19.22 | 35.8 | 0.4065 | 13.4748 | |
| IGClassifier=ViT-B/32, Attribution Target=Predicted Class, Runtime=0.157s2026.04 | 0.2426 | 0.6567 | 39.0279 | 55.42 | 18.97 | 36.45 | 0.4141 | 14.2831 | |
| Bi-Att.Classifier=ViT-B/32, Attribution Target=Predicted Class, Runtime=0.158s2026.04 | 0.2416 | 0.6782 | 34.3086 | 57.26 | 18.87 | 38.39 | 0.4366 | 17.6613 | |
| TISClassifier=ViT-B/32, Attribution Target=Predicted Class, Runtime=0.416s2026.04 | 0.2144 | 0.7181 | 48.3937 | 60.94 | 17.02 | 43.92 | 0.5037 | 9.7104 | |
| AHAClassifier=ViT-B/32, Attribution Target=Predicted Class, Runtime=0.016s2026.04 | 0.1907 | 0.6703 | 39.714 | 56.49 | 14.71 | 41.78 | 0.4796 | 11.5998 | |
| AHAClassifier=ViT-B/32, Attribution Target=Predicted Class, Refinement steps T=3, Runtime=0.212s2026.04 | 0.16 | 0.7224 | 47.8573 | 60.8 | 12.61 | 48.19 | 0.5624 | 7.8339 |