List your company

Visual Question Answering on VQA v2 (test-dev) (Robustness)

71.1Accuracy (Noise)Latent Denoising
69.33269.79170.2570.709Apr 23, 2026

Evaluation results

MethodLinks
Latent DenoisingBackbone=LLaVA+CLIP, Corruption Severity=52026.0471.169.373.373
BaselineBackbone=LLaVA+CLIP, Corruption Severity=52026.0469.467.871.470.9