| { | |
| "dataset_name": "AI Skating Coach - Clean 64-Class", | |
| "version": "1.0", | |
| "description": "Figure skating element recognition with 64 classes including multi-jump combinations", | |
| "total_samples": 5168, | |
| "train_samples": 4161, | |
| "test_samples": 1007, | |
| "split_ratio": "80/20", | |
| "num_classes": 64, | |
| "sequence_length": 150, | |
| "num_keypoints": 17, | |
| "coordinates_per_keypoint": 3, | |
| "duration_seconds": 5.0, | |
| "fps": 30, | |
| "keypoint_format": "COCO 17-point", | |
| "data_characteristics": { | |
| "no_augmentation": true, | |
| "no_class_weights": true, | |
| "no_synthetic_samples": true, | |
| "combinations_preserved": true, | |
| "format": "Pickle (NumPy-compatible, list of arrays)" | |
| }, | |
| "data_sources": [ | |
| "MMFS Dataset: 2D pose from competition videos (4,915 sequences)", | |
| "JSON Motion Capture: Professional 3D mocap (253 sequences)" | |
| ], | |
| "preprocessing_steps": [ | |
| "Format unification (142-marker mocap \u2192 17-keypoint COCO)", | |
| "Temporal uniform sampling to 150 frames", | |
| "Coordinate normalization to [-1, 1]", | |
| "Velocity feature computation", | |
| "80/20 stratified train/test split" | |
| ], | |
| "keypoint_definitions": { | |
| "0-4": "Head (nose, left_eye, right_eye, left_ear, right_ear)", | |
| "5-10": "Upper body (left_shoulder, right_shoulder, left_elbow, right_elbow, left_wrist, right_wrist)", | |
| "11-16": "Lower body (left_hip, right_hip, left_knee, right_knee, left_ankle, right_ankle)" | |
| } | |
| } |