VBVR-Dataset: Very Big Video Reasoning Training Data

VBVR-Dataset: Very Big Video Reasoning Training Data Overview VBVR-Dataset is an unprecedentedly large-scale video reasoning training resource, part of the Very Big Video Reasoning (VBVR) Suite. This release contains the training split: 100 curated reasoning task generators with 1,000,000 video clips (10,000 samples per generator), with each sample consisting of a video, start/end frames, a textual reasoning prompt, and… See the full description on the dataset page:

Source: Hugging Face Hub (Video-Reason/VBVR-Dataset). Metadata imported from the dataset’s Hub tags.