SahasraK/LADBench

LADBench: A Benchmark for Logical Anomaly Detection in Images Large Vision Language Models (VLMs) excel at visual question answering and semantic grounding, but their capacity for autonomous logical reasoning remains underexplored. Existing anomaly benchmarks emphasize visual errors or direct prompting rather than the physical and social common sense needed for open-world deployment. To address this, we introduce LAD-Bench, a benchmark of more than 1,000 curated synthetic images… See the full description on the dataset page: https://huggingface.co/datasets/SahasraK/LADBench.