How fakebaconpancakes/reat---trial works
This repository implements NTU RGB+D skeleton-based human-action recognition, model training, inference, and explainability evaluation. The main workflow prepares skeleton data, converts each sample into geometric and motion features, trains a spatial-GCN/temporal-attention classifier, then evaluates predictions and generates joint-level explanations. Separate proving engines assess explanation faithfulness and semantic alignment. The README is empty, and several central scripts were not sampled; their components are shown where supported by the tree, but uncertain wiring is omitted.
Main entry points: batch_processor.py Batch inference · dataset.py Dataset features · proving_engine_fidelity.py Faithfulness test · train.py Model training
Components 16
Skeleton data
- Data preprocessing
preprocess_data.py - Data splitting
data_splitter.py - Dataset features
utils/dataset.py
Recognition model
- Model training
train.py - Spatial GCN
models/spatial_gcn.py - Temporal attention
models/temporal_brain.py - Model weights
Explainability
- Batch inference
batch_processor.py - Heatmap scaling
utils/xai_extractor.py - Joint heatmaps
- XAI visualizations
visualization_gif.py
Evaluation
- Model evaluation
evaluate.py - Faithfulness test
proving_engine_fidelity.py - Semantic test
proving_engine_semantic.py - Evaluation metrics
Other components
- Skeleton files
Connections 24
- Skeleton files → Data preprocessing: processes
- Data preprocessing → Dataset features: prepares data
- Dataset features → Model training: provides batches
- Model training → Spatial GCN: invokes
- Model training → Temporal attention: invokes
- Model training → Model weights: saves
- Dataset features → Model evaluation: provides data
- Model weights → Model evaluation: loads
- Dataset features → Batch inference: provides samples
- Batch inference → Spatial GCN: invokes
- Batch inference → Temporal attention: invokes
- Batch inference → Model weights: loads
- Batch inference → Heatmap scaling: normalizes heatmaps
- Batch inference → Joint heatmaps: writes
- Batch inference → XAI visualizations: renders
- Dataset features → Faithfulness test: provides samples
- Joint heatmaps → Faithfulness test: reads
- Faithfulness test → Spatial GCN: invokes
- Faithfulness test → Temporal attention: invokes
- Faithfulness test → Model weights: loads
- Faithfulness test → Evaluation metrics: writes
- Dataset features → Semantic test: provides labels
- Joint heatmaps → Semantic test: reads
- Semantic test → Evaluation metrics: writes