--- dataset_info: features: - name: query dtype: string - name: negative_frames list: - name: start_time dtype: float64 - name: end_time dtype: float64 - name: output_path dtype: string - name: frame_path dtype: string - name: positive_frames list: - name: start_time dtype: float64 - name: end_time dtype: float64 - name: output_path dtype: string - name: frame_path dtype: string - name: input_frames dtype: string splits: - name: test num_bytes: 1984070 num_examples: 1800 download_size: 380078 dataset_size: 1984070 configs: - config_name: default data_files: - split: test path: data/test-* --- Original MomentSeeker data ([jsonl file](https://huggingface.co/datasets/VLM2Vec/MMEB-V2/blob/main/video-tasks/data/momentseeker.jsonl)). It contains 1,800 rows and duplicate queries ([discussion](https://github.com/TIGER-AI-Lab/VLM2Vec/issues/123#issuecomment-3141653760)). Check out https://huggingface.co/datasets/VLM2Vec/MomentSeeker for the dedup version.