LabOS LSV Benchmark

LabOS LSV Benchmark LabOS LSV is the benchmark for evaluating video-language models on wet-lab procedural supervision. It contains egocentric, third-person, and multiview laboratory videos paired with protocol-aligned evaluation manifests for monitoring what step is happening, whether the monitored procedure advances, and whether tool clips contain a certain type of error. The package includes the media, benchmark manifests, prompt loaders, inference runner, and report generator… See the full description on the dataset page:

Source: Hugging Face Hub (cong-lab/lsv). Metadata imported from the dataset’s Hub tags.