Back to Blog
SEO

# High-Quality Jogging Pose Dataset – 14,550 Frames Ready for Robot Training

April 1, 20262 min read

We just released our newest **Jogging Pose Dataset** — 61 real-world jogging videos processed with strict HQ filtering.

### Dataset Specifications

- **Videos**: 61 (Pexels source)

- **HQ Frames exported**: **14,550** (incl. horizontal flip + keypoint noise augmentations)

- **Accepted frames (pre-augmentation)**: 4,850

- **Acceptance rate**: **67.81%**

- **Mean Quality Score**: **0.857**

- **Mean landmark visibility**: 0.806

- **Mean motion consistency**: 0.908

**Technical features included:**

- 33 keypoints (MediaPipe BlazePose compatible)

- Gaussian temporal smoothing (window = 5 on x/y/z)

- Hip-center + torso-scale body normalization

- Velocity, acceleration, stride length, and motion consistency metadata

- Full quality report + rejected frames for transparency

### Perfect for:

- Humanoid robot locomotion training

- Imitation learning & reinforcement learning

- Gait analysis & sports biomechanics

- Action recognition models

**Price**: One-time purchase on Gumroad (commercial license).

[Buy Jogging Pose Dataset now →](https://qvision.space/dataset-pricing)