RenShuhuai-Andy / TimeChat

[CVPR 2024] TimeChat: A Time-sensitive Multimodal Large Language Model for Long Video Understanding
https://arxiv.org/abs/2312.02051
BSD 3-Clause "New" or "Revised" License
267 stars 23 forks source link

Subset of YT-Temporal #24

Closed patrick-tssn closed 4 months ago

patrick-tssn commented 4 months ago

Thank you for your efforts. Given the vast scale of the YT-Temporal-1B dataset, and considering that you have sampled a subset of 31.6K entries, would it be possible for you to provide direct access to this specific subset for ease of use? Your assistance is greatly appreciated. Thank you very much.

RenShuhuai-Andy commented 4 months ago

Hi, thanks for your interest.

Unfortunately, we cannot provide direct access to these videos due to huge storage overhead.

You can extract the video_id from instruct_tsg_31.6k_yttemporal.json, e.g., -QVtAHb90Vc in "video": "yttemporal180m/videos/-QVtAHb90Vc.mp4", than use the command os.system(f"yt-dlp -f 'bv*[ext=mp4]+ba[ext=m4a]/b[ext=mp4] / bv*+ba/b' -o {os.path.join(folder, 'videos', video_id)} www.youtube.com/watch?v={video_id}") to download them.