Ready-made dataset

High-Quality Walking Pose Dataset — 22,923 HQ Frames (100 videos)

Walking / gait-oriented bulk export for pose estimation, locomotion research, and robotics-style training pipelines.

Built with the Quality Vision Motion Dataset Engine (layer11_pose): 100 diverse source clips, strict HQ gating, temporal smoothing, and standard augmentations.

One-time: $370Source videos: 100Rows exported (incl. augments): 22,923HQ accepted (pre-aug): 7,641

Dataset overview

  • Action label: walking (bulk job bulk_100_videos_f3b26484)
  • Exported rows (incl. augmentations): 22,923 in merged data.jsonl
  • HQ frames (pre-augmentation): 7,641 accepted
  • Mean quality (accepted): ~0.863 frame-quality score
  • Mean landmark visibility (accepted): ~0.810
  • Stride: 3 · Gaussian smoothing: window 5 (x/y/z)

Technical stack

  • Pose: BlazePose-style landmarks (normalized), optional Layer 1.1 scene stats
  • Augmentations: horizontal flip + keypoint noise (see manifest.json)
  • Deliverables: data.jsonl, per_video/, bulk_combined_manifest.json, QA JSON

Structured layout

  • data.jsonl — merged training file
  • per_video/*.jsonl — per-source splits
  • features.json, global_stats.json, export_quality_report.json

Free samples

Validate parsing on GitHub: QualityVision-Motion-Dataset-Samples.

Purchase & licensing

  • Price: $370 one-time (Gumroad)
  • Deliverable: engineered exports (JSONL + metadata) — not raw source videos
  • Contact: info@qvision.space
Buy on Gumroad $370 Back to pricing