Feedback Request for Free high-quality walking pose dataset sample ( frames, HQ filtered + temporal smoothing) – seeking your thoughts for motion AI / robotics
#1
by Alaaharoun - opened
Hi ; I have a small samples and need feedback for early samples, I'm sharing a small walking pose dataset processed with:
- High-Quality filtering (min avg visibility 0.65, frame quality 0.45+)
- Temporal Gaussian smoothing on accepted frames
- Body-part diagnostics (face / upper / arms / lower_body)
- Per-frame and sequence-level quality reports
Sample stats (after filtering): - Accepted frames: 98 (57% acceptance rate)
- Mean avg landmark visibility: 0.764
- Mean lower-body visibility: 0.889
- Includes: data.jsonl, features.json, quality_report, manifest, low_quality_frames (for transparency)
You can download the sample ZIP here: GitHub https://github.com/Alaaharoun/QualityVision-Motion-Dataset-Samples
What I'm looking for:
Is the visibility and motion consistency good enough for your use case (imitation learning, sports analysis, robotics, etc.)?
What would you improve? (e.g. longer sequences, more actions, 3D lifting, better smoothing...)
Any missing metrics or features you'd like to see?
Would you be interested in purchasing such datasets? We can extract and process any type of movement you need (walking, running, jogging, turning, sports movements, carrying objects, etc.) according to your specific requirements. We also offer custom processing: simply send us your raw videos, and we will deliver a clean, high-quality structured pose dataset with full quality reports and specifications.
This is completely free for feedback. No strings attached — just honest opinions to make the dataset better.
Thanks in advance! I'll reply to all comments.