Add walking sample + dataset card

README.md

@@ -1,3 +1,64 @@
 ---
 license: cc-by-nc-4.0
+language:
+- en
+tags:
+- pose-estimation
+- walking
+- 2d-pose
+- mediapipe
+- motion-dataset
+- quality-vision
+pretty_name: QualityVision Walking Sample (2D pose)
+size_categories:
+- n<1K
 ---
+
+# QualityVision Walking Sample
+
+**High-quality 2D pose estimation** sample for **walking**, exported from the [Quality Vision Motion Dataset Engine](https://github.com/Alaaharoun/QualityVision-Motion-Dataset-Samples) pipeline: **HQ frame filtering**, optional **temporal smoothing** on accepted frames, and **sequence diagnostics** in `features.json`.
+
+## Files
+
+| File | Description |
+|------|-------------|
+| `data.jsonl` | One JSON object per line: normalized image-space keypoints, timestamps, optional Layer 1.1 metadata. |
+| `features.json` | Sequence-level aggregates (visibility, hip motion, `motion_consistency`). |
+| `global_stats.json` | Job-level roll-up for this clip. |
+| `manifest.json` | Export metadata (format compatible with the Motion Dataset Engine). |
+| `export_quality_report.json` | Pilot-oriented quality summary. |
+
+## Usage
+
+```python
+from datasets import load_dataset
+
+# If you load as JSON lines:
+ds = load_dataset("Alaaharoun/QualityVision-walking-sample", data_files="data.jsonl", split="train")
+```
+
+Or download raw files:
+
+```python
+from huggingface_hub import hf_hub_download
+path = hf_hub_download(
+    repo_id="Alaaharoun/QualityVision-walking-sample",
+    filename="data.jsonl",
+    repo_type="dataset",
+)
+```
+
+## License
+
+**CC BY-NC 4.0** — see repository `LICENSE` on [GitHub](https://github.com/Alaaharoun/QualityVision-Motion-Dataset-Samples). Commercial use outside NC terms requires a separate agreement.
+
+## Citation
+
+```bibtex
+@misc{qualityvision_walking_sample,
+  title = {QualityVision Walking Sample (2D Pose)},
+  author = {Alaaharoun},
+  howpublished = {\url{https://huggingface.co/datasets/Alaaharoun/QualityVision-walking-sample}},
+  year = {2026},
+}
+```

README_sample.md

@@ -0,0 +1,19 @@
+# `walking_sample_v1`
+
+| File | Description |
+|------|-------------|
+| `data.jsonl` | 80 consecutive frames, one JSON object per line (primary person `keypoints`, optional `layer11`, `source_video_fps`, etc.). |
+| `features.json` | Sequence-level aggregates (visibility, hip motion, `motion_consistency`, stride proxies). |
+| `global_stats.json` | Job-level roll-up for this clip. |
+| `manifest.json` | Metadata aligned with the Motion Dataset Engine export format. |
+| `export_quality_report.json` | Pilot-oriented summary + pointers to formulas. |
+
+**Coordinate system:** normalized image coordinates (typically ~0–1 for x/y), z relative depth as exported by the engine.
+
+**Not included:** raw video, full ZIP checksums, or rejected frames — this repo is intentionally small for GitHub hosting and feedback collection.
+
+To regenerate a pose preview image, run from the repository root:
+
+```bash
+python scripts/render_pose_preview.py
+```

export_quality_report.json

@@ -0,0 +1,20 @@
+{
+  "sample_id": "walking_sample_v1",
+  "purpose": "free_public_pilot",
+  "frames_included": 80,
+  "engine": "Quality Vision Motion Dataset Engine",
+  "pipeline_summary": "2D pose (MediaPipe-style landmarks), HQ visibility/motion gate, body-part diagnostics; parent job used optional temporal smoothing on accepted frames.",
+  "sequence_metrics": {
+    "quality_score": 0.6328,
+    "mean_avg_landmark_visibility": 0.770697,
+    "motion_consistency": 0.494943,
+    "motion_consistency_method": "blend_smoothed_cv_mad_v1",
+    "short_sequence_motion_warning": false
+  },
+  "documentation": {
+    "repository": "https://github.com/Alaaharoun/QualityVision-Motion-Dataset-Samples",
+    "per_frame_scores": "See Motion Dataset Engine: weighted visibility + lower-body + motion_local (action-aware nudge).",
+    "sequence_scores": "See features.json sequence block; motion_consistency blends smoothed step-length CV with a MAD/median robust term."
+  },
+  "license": "CC-BY-NC-4.0 for this sample distribution; commercial use requires a separate agreement (see docs/EULA.md)."
+}

features.json

@@ -0,0 +1,33 @@
+{
+  "schema_version": "1.2.1",
+  "job_id": "qv_walking_sample_v1",
+  "sequence": {
+    "frame_count": 80,
+    "action_label": "walking",
+    "avg_landmark_visibility": 0.770697,
+    "hip_center_displacement_mean_norm": 0.014724,
+    "hip_center_path_sum_norm": 1.163219,
+    "inter_frame_steps": 79,
+    "velocity_proxy_available": true,
+    "velocity_mean": 0.014724,
+    "velocity_std": 0.010937,
+    "velocity_cv_raw": 0.742789,
+    "hip_velocity_smooth_window": 3,
+    "velocity_mean_smoothed": 0.014721,
+    "velocity_std_smoothed": 0.008312,
+    "velocity_cv_smoothed": 0.564646,
+    "velocity_cv_mad": 0.394393,
+    "motion_consistency_cv_smoothed": 0.435354,
+    "motion_consistency_mad": 0.605607,
+    "motion_consistency_blend_cv_weight": 0.65,
+    "short_sequence_motion_warning": false,
+    "acceleration_mean": 0.008273,
+    "avg_stride_length": 0.038663,
+    "body_height_norm": 0.351658,
+    "arm_swing_amplitude": 0.086108,
+    "step_frequency": 24.6875,
+    "motion_consistency": 0.494943,
+    "motion_consistency_method": "blend_smoothed_cv_mad_v1"
+  },
+  "notes": "norm = image-normalized coordinates (0–1); x/y clamped at export. Hip displacement is Euclidean distance in normalized space between consecutive exported frames. motion_consistency blends (1) stability of CV on a short moving-average of step lengths (reduces pose jitter) and (2) a robust MAD/median term (less harsh than raw std/mean on periodic gait). Raw velocity_cv can still be high for natural walking; prefer smoothed + MAD components. short_sequence_motion_warning=true when inter_frame_steps is below the configured threshold — sequence-level quality is less reliable for imitation learning. For higher motion_consistency in exports: lower stride, longer clips, walking_focused preset, temporal smoothing on accepted frames. stride_length = mean horizontal ankle separation per frame. body_height_norm = mean |nose.y − mid(ankles).y|. step_frequency ≈ hip displacement steps per second of source timeline."
+}

global_stats.json

@@ -0,0 +1,21 @@
+{
+  "schema_version": "1.2.1",
+  "job_id": "qv_walking_sample_v1",
+  "video_filename": "walking_uhd_sample.mp4",
+  "fps": 25.0,
+  "stride": 3,
+  "exported_frames": 80,
+  "max_frame_id": 79,
+  "estimated_source_duration_sec": 3.2,
+  "fps_mean": 25.0,
+  "avg_sequence_length": 80,
+  "action_distribution": {
+    "walking": 80
+  },
+  "quality_score": 0.6328,
+  "artifacts": {
+    "per_frame_records": "data.jsonl",
+    "sequence_features": "features.json",
+    "job_manifest": "manifest.json"
+  }
+}

manifest.json

@@ -0,0 +1,22 @@
+{
+  "schema_version": "1.0.0",
+  "job_id": "qv_walking_sample_v1",
+  "video_filename": "walking_uhd_sample.mp4",
+  "fps": 25.0,
+  "frame_count_exported": 80,
+  "stride": 3,
+  "action_label": "walking",
+  "augmentations_applied": [],
+  "license": "CC-BY-NC-4.0",
+  "consent": true,
+  "notes": "Public pilot sample (80 frames). Produced with the Quality Vision Motion Dataset Engine: HQ frame filter, optional temporal smoothing on accepted frames in the parent export, body-part diagnostics. Filename is illustrative; coordinates are image-normalized. See repository README and docs/EULA.md.",
+  "dataset_format": "pose_estimation_sequence_dataset",
+  "export_manifest_version": "1.1",
+  "bulk_export": false,
+  "source_video_filenames": ["walking_uhd_sample.mp4"],
+  "hq_export": true,
+  "quality_filter": {
+    "enabled": true,
+    "description": "Subset taken from a longer HQ-filtered export; frames renumbered 0–79 for this sample."
+  }
+}