zachz/Vinoground-PC

--- dataset_info: features: - name: video dtype: video - name: major dtype: string splits: - name: test num_examples: 1000 configs: - config_name: default data_files: - split: test path: data/* license: apache-2.0 task_categories: - video-classification language: - en pretty_name: Vinoground Pair Classification --- # Vinoground-PC Video pair classification dataset from [Vinoground](https://arxiv.org/abs/2410.02763) (Zhang et al., 2024). Each row contains a video clip and its major category (action/viewpoint/object). The MTEB task generates same-category (positive) and different-category (negative) pairs to test whether embeddings capture temporal/compositional differences. Source: `HanSolo9682/Vinoground` — 500 counterfactual pairs, 1000 individual clips.