How ksyegorov/mcd_rppg works
MCD-rPPG is a research codebase for reproducing remote photoplethysmography experiments: prepare face-video inputs, train or run rPPG models, then evaluate pulse signals and heart rate across datasets. The README also describes a multi-task model for health biomarkers; it is represented as a documented capability, but its end-to-end implementation is not established by the sampled source. The graph uses the documented training notebooks as entry points. Source excerpts support the training, preprocessing, and evaluation links shown; unsampled notebook and model wiring is not inferred.
Main entry points: train.py Fold training and testing · models Neural model zoo · preprocessing Dataset preprocessing · __init__.py Model registry
Components 17
Video preparation
- Dataset preprocessing
preprocessing - Face and landmarks
rppglib/face_utils.py
Data and signals
- Video loading
rppglib/data_utils.py - Training dataset
rppglib/train.py - Windowed dataset API
rppglib/dataset.py - Signal processing
rppglib/processing.py
rPPG models
- Model registry
rppglib/models/__init__.py - Neural model zoo
rppglib/models - Unsupervised methods
rppglib/unsupervised.py
Training and evaluation
- Model training notebooks
train_SCNN_8roi_mcd_rppg.ipynb - Other training experiments
train_POS.ipynb - Fold training and testing
rppglib/train.py - Inference timing
Inference_time.ipynb
Other components
- Researcher
- Video and PPG datasets
- PPG and heart rate
- Health biomarkers
Connections 19
- Researcher → Model training notebooks: runs experiment
- Researcher → Other training experiments: runs experiment
- Researcher → Inference timing: runs inference
- Video and PPG datasets → Dataset preprocessing: provides videos
- Dataset preprocessing → Video loading: loads video
- Dataset preprocessing → Face and landmarks: extracts faces
- Model training notebooks → Fold training and testing: starts training
- Other training experiments → Fold training and testing: starts training
- Video and PPG datasets → Fold training and testing: provides metadata
- Fold training and testing → Model registry: selects model
- Model registry → Neural model zoo: resolves model
- Fold training and testing → Training dataset: builds loaders
- Training dataset → Signal processing: filters PPG
- Fold training and testing → Signal processing: calculates metrics
- Fold training and testing → PPG and heart rate: reports PPG and HR
- Inference timing → Neural model zoo: benchmarks models
- Neural model zoo → PPG and heart rate: predicts pulse
- Neural model zoo → Health biomarkers: estimates biomarkers
- Unsupervised methods → PPG and heart rate: extracts pulse