VatsaDev/raw-vid-danbooru

This is a raw example dataset

• this was designed to prove that my pipelines work and scale (took a day and half, but it works)
• its currently at 360p as that takes half the time of 720p, the real dataset would be less blurry
• Its currently more japanese than english (at least as of the 600th), but I've found some higher quality datasets (better playlists) and will be using those from this point, that should probably get me 60-80k samples

Current dataset

• current size at ~1100 ex.
• eventually scale to the image-danbooru size (~5M)

The file structure is just:

<video_id>\
... metadata.jsonl
... clip1.mp4
... clip2.mp4
... etc

metadata contains the text per clip