IntuitivePhysics

WorldBench WorldBench is a new benchmark designed to evaluate the physical understanding and prediction of modern world models and vision-language models. There are two components: Video based: This is a benchmark designed to evaluate video-to-video world foundation models such as Cosmos. It consists of videos 132 frames long of 425 simulated scenes with RGB, Normals, Depth, Flow, and Segmentations. Text based: This is a subset which adds text-based questions to 181 videos from… See the full description on the dataset page:

Source: Hugging Face Hub (worldbenchmark/IntuitivePhysics). Metadata imported from the dataset’s Hub tags.