Training Data Exposure Detection on COCO (train vs Non-training)
87.2AUCGradAudit
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GradAuditModel=BLIP-ITM, Architecture=Retrieval, Params=200M, Training Stage=Pretrain2026.06 | 87.2 | 49.2 | |
| GradAuditModel=MiniGPT-v2, Architecture=Generation, Params=8.2B, Training Stage=Pretrain2026.06 | 63.7 | 21.5 | |
| GradNormModel=BLIP-ITM, Architecture=Retrieval, Params=200M, Training Stage=Pretrain2026.06 | 59.3 | 10.9 | |
| Min-KModel=BLIP-ITM, Architecture=Retrieval, Params=200M, Training Stage=Pretrain2026.06 | 55.6 | 6.7 | |
| M4IModel=BLIP-ITM, Architecture=Retrieval, Params=200M, Training Stage=Pretrain2026.06 | 55.5 | 6 | |
| EntropyModel=BLIP-ITM, Architecture=Retrieval, Params=200M, Training Stage=Pretrain2026.06 | 55.3 | 6.9 | |
| Min-K++Model=BLIP-ITM, Architecture=Retrieval, Params=200M, Training Stage=Pretrain2026.06 | 54.3 | 6.38 | |
| DC-PDDModel=BLIP-ITM, Architecture=Retrieval, Params=200M, Training Stage=Pretrain2026.06 | 54.3 | 1.9 | |
| ZlibModel=BLIP-ITM, Architecture=Retrieval, Params=200M, Training Stage=Pretrain2026.06 | 52 | 3.1 | |
| LossModel=BLIP-ITM, Architecture=Retrieval, Params=200M, Training Stage=Pretrain2026.06 | 51.3 | 5.41 | |
| M4IModel=MiniGPT-v2, Architecture=Generation, Params=8.2B, Training Stage=Pretrain2026.06 | 51.3 | 6 | |
| ModRenyiModel=BLIP-ITM, Architecture=Retrieval, Params=200M, Training Stage=Pretrain2026.06 | 51.2 | 6 | |
| Min-K++Model=MiniGPT-v2, Architecture=Generation, Params=8.2B, Training Stage=Pretrain2026.06 | 51.1 | 6.12 | |
| ModRenyiModel=MiniGPT-v2, Architecture=Generation, Params=8.2B, Training Stage=Pretrain2026.06 | 50.8 | 5.2 | |
| ZlibModel=MiniGPT-v2, Architecture=Generation, Params=8.2B, Training Stage=Pretrain2026.06 | 50.7 | 2.81 | |
| SimilarityModel=BLIP-ITM, Architecture=Retrieval, Params=200M, Training Stage=Pretrain2026.06 | 50.7 | 5.2 | |
| LossModel=MiniGPT-v2, Architecture=Generation, Params=8.2B, Training Stage=Pretrain2026.06 | 50.6 | 3.3 | |
| Min-KModel=MiniGPT-v2, Architecture=Generation, Params=8.2B, Training Stage=Pretrain2026.06 | 50.4 | 4.4 | |
| GradNormModel=MiniGPT-v2, Architecture=Generation, Params=8.2B, Training Stage=Pretrain2026.06 | 50.4 | 5.5 | |
| DC-PDDModel=MiniGPT-v2, Architecture=Generation, Params=8.2B, Training Stage=Pretrain2026.06 | 50.1 | 5.1 | |
| EntropyModel=MiniGPT-v2, Architecture=Generation, Params=8.2B, Training Stage=Pretrain2026.06 | 48.6 | 4.31 | |
| SimilarityModel=MiniGPT-v2, Architecture=Generation, Params=8.2B, Training Stage=Pretrain2026.06 | 48.3 | 5.67 |