Multi-modal Model Evaluation on MMStar (Corruption Robustness)
47.6Noise ScoreLatent Denoising
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Latent DenoisingArchitecture=Qwen-2.5-VL, Severity=52026.04 | 47.6 | 42.7 | 56.7 | 51.8 | |
| BaselineArchitecture=Qwen-2.5-VL, Severity=52026.04 | 44.8 | 42 | 53.6 | 50.3 | |
| Latent DenoisingBackbone=LLaVA+CLIP, Corruption Severity=52026.04 | 35.3 | 36.1 | 36.8 | 35.8 | |
| BaselineBackbone=LLaVA+CLIP, Corruption Severity=52026.04 | 29.5 | 30.8 | 32.5 | 30.6 |