Neuron Interpretation on ImageNet CoSy benchmark avgpool layer 1k
0.97AUCLINE
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LINEModel Architecture=ResNet50, Layer=avgpool2026.04 | 0.97 | 5.1 | |
| LINEModel Architecture=ResNet18, Layer=avgpool2026.04 | 0.96 | 4.61 | |
| LINEModel Architecture=ViT-B/16, Layer=encoder2026.04 | 0.94 | 2 | |
| CLIP-DissectModel Architecture=ResNet18, Layer=avgpool2026.04 | 0.91 | 3.78 | |
| INVERTModel Architecture=ResNet50, Layer=avgpool2026.04 | 0.88 | 2.53 | |
| CLIP-DissectModel Architecture=ResNet50, Layer=avgpool2026.04 | 0.87 | 3.59 | |
| INVERTModel Architecture=ResNet18, Layer=avgpool2026.04 | 0.83 | 2.08 | |
| DnDModel Architecture=ResNet50, Layer=avgpool2026.04 | 0.76 | 2.03 | |
| INVERTModel Architecture=ViT-B/16, Layer=encoder2026.04 | 0.76 | 0.96 | |
| DnDModel Architecture=ResNet18, Layer=avgpool2026.04 | 0.73 | 1.79 | |
| CLIP-DissectModel Architecture=ViT-B/16, Layer=encoder2026.04 | 0.69 | 0.78 | |
| DnDModel Architecture=ViT-B/16, Layer=encoder2026.04 | 0.56 | 0.27 |