Text-to-Music Generation on MusicCaps (Poisoning Robustness Evaluation)
0.33CLAP Similarity (Benign, User Question)dual-layer caption poisoning attack
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| dual-layer caption poisoning attackTTM Model=MusicGen, Top-k=32026.05 | 0.33 | 0.28 | 0.28 | 0.48 | |
| dual-layer caption poisoning attackTTM Model=MusicGen, Top-k=52026.05 | 0.31 | 0.25 | 0.27 | 0.45 | |
| dual-layer caption poisoning attackTTM Model=MusicGen, Top-k=102026.05 | 0.29 | 0.21 | 0.3 | 0.41 |