Latent Training Semantics Recovery on ImageNet 1K (train)
73Semantic Similarity ScoreCLIPDissect
Evaluation results
| Method | Links | |
|---|---|---|
| CLIPDissectAccess to model weights (W)=true, Access to training data (D)=true, Backbone=RN502026.03 | 73 | |
| UNBOXAccess to model weights (W)=false, Access to training data (D)=false, Backbone=RN502026.03 | 63 | |
| DEXTERAccess to model weights (W)=true, Access to training data (D)=false, Backbone=RN502026.03 | 49 | |
| UNBOXAccess to model weights (W)=false, Access to training data (D)=false, Backbone=ViT2026.03 | 45 | |
| WWWAccess to model weights (W)=true, Access to training data (D)=true, Backbone=RN502026.03 | 43 | |
| DEXTERAccess to model weights (W)=true, Access to training data (D)=false, Backbone=ViT2026.03 | 34 | |
| CLIPDissectAccess to model weights (W)=true, Access to training data (D)=true, Backbone=ViT2026.03 | 30 | |
| WWWAccess to model weights (W)=true, Access to training data (D)=true, Backbone=ViT2026.03 | 27 | |
| RandomAccess to model weights (W)=false, Access to training data (D)=false, Backbone=RN502026.03 | 3 | |
| RandomAccess to model weights (W)=false, Access to training data (D)=false, Backbone=ViT2026.03 | 3 |