How mehmet-sarikaya/skeleton-based-basketball-action-recognition works
This is a Python application for recognizing basketball gestures from video using pose keypoints and temporal classification. Its main workflow is to extract and label skeleton data, prepare fixed-length windows, train a selected model with grouped validation, then classify video or camera input in real time. The README also describes augmentation and model comparisons. Edges follow shown imports and calls or README-described menu capabilities; unsampled implementation details are not inferred.
Main entry points: BasketballGestureRecognitionApp.py App menu · ModelCreator.py Model selection · ModelTester.py Video and camera inference · KeypointDatasetProcessor.py Window preparation
Components 16
Application workflow
- App menu
BasketballGestureRecognitionApp.py - Training and validation
ModelTrainer.py
Pose data preparation
- Dataset extraction
KeypointDatasetCreator.py - Video sampling
VideoProcessor.py - YOLO pose
- Window preparation
KeypointDatasetProcessor.py - Skeleton augmentation
Augmenter.py - Skeleton visualization
SkeletonAnimator.py
Gesture models
- Model selection
ModelCreator.py - ST-GCN layers
Graph_Model_Creator.py - ST-GCN classifier
stgcn_model.py - Skeleton graph
stgcn_graph.py - TCN attention model
tchm_model.py
Inference
- Video and camera inference
ModelTester.py - Gesture labels
Other components
- Operator
Connections 19
- Operator → App menu: chooses mode
README.md - App menu → Dataset extraction: starts extraction
BasketballGestureRecognitionApp.py - Dataset extraction → Video sampling: samples video
KeypointDatasetCreator.py - Dataset extraction → YOLO pose: extracts poses
KeypointDatasetCreator.py - App menu → Window preparation: loads dataset
BasketballGestureRecognitionApp.py - Window preparation → Skeleton augmentation: augments windows
KeypointDatasetProcessor.py - Window preparation → Skeleton visualization: visualizes windows
KeypointDatasetProcessor.py - App menu → Training and validation: starts training
README.md - Training and validation → Window preparation: uses prepared data
ModelTrainer.py - Training and validation → Model selection: requests model
ModelTrainer.py - App menu → Model selection: selects architecture
BasketballGestureRecognitionApp.py - Model selection → ST-GCN layers: builds ST-GCN
ModelCreator.py - Model selection → ST-GCN classifier: selects classifier
ModelCreator.py - ST-GCN classifier → Skeleton graph: constructs graph
stgcn_model.py - Model selection → TCN attention model: builds TCN model
ModelCreator.py - App menu → Video and camera inference: starts inference
README.md - Video and camera inference → Video sampling: processes frames
ModelTester.py - Video and camera inference → YOLO pose: tracks poses
ModelTester.py - Video and camera inference → Gesture labels: displays classifications
ModelTester.py