Robustness Evaluation on VizWiz
70.9AccuracyLatent Denoising
Evaluation Results
| Method | Links | |
|---|---|---|
| Latent DenoisingArchitecture=Qwen-2.5-VL2026.04 | 70.9 | |
| BaselineArchitecture=Qwen-2.5-VL2026.04 | 68.9 | |
| Latent DenoisingArchitecture=LLaVA+SigLIP2026.04 | 57.8 | |
| Latent DenoisingArchitecture=LLaVA+CLIP2026.04 | 54.8 | |
| BaselineArchitecture=LLaVA+SigLIP2026.04 | 54.4 | |
| BaselineArchitecture=LLaVA+CLIP2026.04 | 54 |