VPoS

VPoS-Bench: Video Pointing and Segmentation Benchmark VPoS-Bench is a challenging out-of-distribution benchmark designed to evaluate the spatio-temporal pointing and reasoning capabilities of video-language models. It covers a diverse set of five real-world application domains, with fine-grained point-level and segmentation annotations that enable robust evaluation of multimodal models under realistic, temporally complex scenarios. Webpage: VideoMolmo Paper: VideoMolmo:… See the full description on the dataset page: https://huggingface.co/datasets/ghazishazan/VPoS.