Curated Dataset Structure

Once a Dataset has been curated, you can download a ZIP file of the curated dataset or view it in your S3 storage location. The contents of a curated dataset is described below.

File/Folder Name	Description
clips	Directory containing the curated clips
iv2_embd	Directory containing the embeddings for the clips
metas	Directory contains the captions for each clip. The clip and captions share the same uuid file in the file names.
previews	Directory containing the webp previews of the curated clips
processed_clip_chunks	Directory containing processed video file segment names
processed_videos	Directory containing processed video file names
summary.json	JSON file containing detailed metadata summary of videos processed, clips generated, and some statistics, along with video/clip metadata
v0	Directory containing the same captions as metas, but inside a single file for easy processing (i.e. v0/all_window_captions.json)