Image Classification on Saliency-Bench
65.51Top-1 AccuracyLIMA (Ours)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| LIMA (Ours)Human Prior=Masks, Model=CLIP (Radford et al., 2021), Attributions=LIMA2026.01 | 65.51 | 82.64 | 56.48 | 91.92 | 750 | |
| XILHuman Prior=Masks, Model=CLIP (Radford et al., 2021), Attributions=Grad-ECLIP2026.01 | 64 | 78.91 | 53.27 | 90.45 | 121 | |
| MEGLHuman Prior=Masks, Model=CLIP (Radford et al., 2021), Attributions=Grad-ECLIP2026.01 | 63.54 | 81.8 | 53.18 | 90.04 | 148 | |
| Fine-tuningHuman Prior=Masks, Model=CLIP (Radford et al., 2021), Attributions=-2026.01 | 60.76 | 78.47 | 52.31 | 90.44 | 43 | |
| RRRHuman Prior=Masks, Model=CLIP (Radford et al., 2021), Attributions=Input Gradient2026.01 | 60.3 | 78.21 | 52.53 | 89.43 | 147 | |
| LIMA (Ours)Human Prior=Masks, Model=ViT (base) (Dosovitskiy et al., 2021), Attributions=LIMA2026.01 | 56.94 | 76.39 | 54.63 | 85.19 | 242 | |
| LIMA (Ours)Human Prior=Masks, Model=ResNet-101 (He et al., 2016), Attributions=LIMA2026.01 | 55.9 | 76.62 | 69.84 | 87.82 | 104 | |
| XILHuman Prior=Masks, Model=ResNet-101 (He et al., 2016), Attributions=Grad-CAM2026.01 | 55.21 | 76.16 | 67.25 | 86.79 | 41 | |
| Fine-tuningHuman Prior=Masks, Model=ResNet-101 (He et al., 2016), Attributions=-2026.01 | 54.98 | 75.69 | 62.35 | 76.94 | 35 | |
| RRRHuman Prior=Masks, Model=ResNet-101 (He et al., 2016), Attributions=Input Gradient2026.01 | 54.98 | 76.04 | 60.76 | 78.57 | 56 | |
| MEGLHuman Prior=Masks, Model=ResNet-101 (He et al., 2016), Attributions=Grad-CAM2026.01 | 54.51 | 76.62 | 63.15 | 83.44 | 47 | |
| RRRHuman Prior=Masks, Model=ViT (base) (Dosovitskiy et al., 2021), Attributions=Input Gradient2026.01 | 53.7 | 75.12 | 45.09 | 65.3 | 123 | |
| MEGLHuman Prior=Masks, Model=ViT (base) (Dosovitskiy et al., 2021), Attributions=Grad-ECLIP2026.01 | 53.59 | 73.38 | 51.45 | 82.42 | 111 | |
| Fine-tuningHuman Prior=Masks, Model=ViT (base) (Dosovitskiy et al., 2021), Attributions=-2026.01 | 51.5 | 73.5 | 43.63 | 77.86 | 37 | |
| XILHuman Prior=Masks, Model=ViT (base) (Dosovitskiy et al., 2021), Attributions=Grad-ECLIP2026.01 | 51.39 | 69.68 | 43.97 | 80.87 | 107 |