Abk802/ORBIT

ORBIT: An Object Property Reasoning Benchmark for Visual Inference Tasks Inspired by human object categorization, this repository introduces ORBIT, a comprehensive benchmark designed to evaluate the abilities of Vision–Language Models (VLMs) to reason about abstract object properties. ORBIT spans four object property dimensions (physical, taxonomic, functional, relational), three levels of reasoning complexity (direct recognition, property inference, counterfactual reasoning), and three… See the full description on the dataset page: https://huggingface.co/datasets/Abk802/ORBIT.