List your company

Visual Question Answering on GQA (Corruption Robustness: Noise, Blur, Weather, Digital)

60.4Accuracy (Noise)Latent Denoising
46.77650.31353.8557.387Apr 23, 2026

Evaluation results

MethodLinks
Latent DenoisingBackbone=LLaVA+CLIP, Corruption Severity=52026.0460.459.76261.6
BaselineBackbone=LLaVA+CLIP, Corruption Severity=52026.0457.456.45958.4
Latent DenoisingArchitecture=Qwen-2.5-VL, Severity=52026.0453.757.962.361
BaselineArchitecture=Qwen-2.5-VL, Severity=52026.0447.351.154.752.6