Datasets:

Bmingg
/

T1_pseudo_gt

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

T1 pseudo-GT segments

Pseudo ground truth, not real ground truth. Segment boundaries and captions are a dummy annotation for pipeline testing.

20 egocentric iPhone recordings (HoangVu, task codes 00300, 00301, 10300, 10301, 5 each).
videos/<session_uid>.mp4: primary video, 1920x1440, 30 fps.
annotations.jsonl: one line per video.
- boundaries_s: boundary times in seconds on the video's own clock.
- task_ranges: [start, end] of the task part of the video; segments cover only this.
- segments: {idx, start_s, end_s, caption}; captions are model-generated (qwen3.7-flash), not QC'd.
- cut_level: granularity label from the cutting tool (L4 or L3.5).