Concept Extraction Consistency on ImageNet
0.84MPPCVan-SAE
Evaluation results
| Method | Links | |
|---|---|---|
| Van-SAEBackbone Model=CLIP2025.12 | 0.84 | |
| DeleuzianBackbone Model=CLIP2025.12 | 0.821 | |
| DeleuzianBackbone Model=DinoV22025.12 | 0.789 | |
| Tk-SAEBackbone Model=CLIP2025.12 | 0.757 | |
| Van-SAEBackbone Model=DinoV22025.12 | 0.603 | |
| Tk-SAEBackbone Model=DinoV22025.12 | 0.588 | |
| ICABackbone Model=CLIP2025.12 | 0.449 | |
| Gat-SAEBackbone Model=CLIP2025.12 | 0.346 | |
| JR-SAEBackbone Model=CLIP2025.12 | 0.341 | |
| JR-SAEBackbone Model=DinoV22025.12 | 0.272 | |
| ICABackbone Model=DinoV22025.12 | 0.264 | |
| Gat-SAEBackbone Model=DinoV22025.12 | 0.264 | |
| Mat-SAEBackbone Model=CLIP2025.12 | 0.225 | |
| Mat-SAEBackbone Model=DinoV22025.12 | 0.201 |