How pratik-337/heart-pumping-efficiency-estimator works
EchoEF is a research prototype that estimates left-ventricular ejection fraction from echocardiogram videos and studies accuracy versus frame count, compute cost, and explanation quality. Its main workflow verifies and preprocesses EchoNet-Dynamic data, trains configured R(2+1)D-18 regressors, predicts on evaluation splits, and analyzes results; a local Gradio demo offers video inference. The README also documents cost/memory probes, XAI, and result reporting. Several links for unsampled CLI subsystems are documented rather than source-verified, so those edges cite the README.
Main entry points: dataset.py Clip dataset · pipeline.py Explanation pipeline · trainer.py Run trainer · inference.py Demo inference
Components 23
Data preparation
- Processed cache
- Dataset verification
scripts/verify_dataset.py - Video preprocessing
src/echoef/data/preprocess.py - Tracing data
src/echoef/data/tracings.py - Clip dataset
src/echoef/data/dataset.py - Clip sampling
src/echoef/data/sampling.py
Training and inference
- Experiment configuration
src/echoef/config.py - EF regressor
src/echoef/models/factory.py - Training steps
src/echoef/train/steps.py - Run trainer
src/echoef/train/trainer.py - Training sweep
src/echoef/train/sweep.py
Evaluation and reporting
- Split evaluation
src/echoef/eval/evaluation.py - Video predictions
src/echoef/eval/predict.py - Cost benchmark
src/echoef/eval/cost.py - Results aggregation
src/echoef/eval/aggregate.py - Figures and tables
src/echoef/eval/figures.py
Explanations
- Explanation pipeline
src/echoef/xai/pipeline.py - Attribution methods
src/echoef/xai/attribution.py - Explanation rendering
src/echoef/xai/render.py
Interactive demo
- Local demo
app/app.py - Demo inference
app/inference.py
Other components
- Researcher
- EchoNet videos
Connections 31
- Researcher → Dataset verification: verifies data
README.md - EchoNet videos → Dataset verification: provides videos
README.md - Dataset verification → Video preprocessing: then preprocesses
README.md - Video preprocessing → Tracing data: uses tracings
src/echoef/data/preprocess.py - Video preprocessing → Processed cache: writes cache
README.md - Processed cache → Clip dataset: supplies frames
src/echoef/data/dataset.py - Clip dataset → Clip sampling: uses clip starts
src/echoef/data/dataset.py - Researcher → Training sweep: starts training
README.md - Training sweep → Run trainer: runs configurations
src/echoef/train/sweep.py - Run trainer → Clip dataset: loads batches
src/echoef/train/trainer.py - Run trainer → Training steps: executes steps
src/echoef/train/trainer.py - Run trainer → EF regressor: builds model
src/echoef/train/trainer.py - Researcher → Split evaluation: requests evaluation
README.md - Split evaluation → Clip dataset: loads split
src/echoef/eval/evaluation.py - Split evaluation → Video predictions: generates predictions
src/echoef/eval/evaluation.py - Video predictions → Clip sampling: selects clips
src/echoef/eval/predict.py - Video predictions → Clip dataset: loads clips
src/echoef/eval/predict.py - Researcher → Cost benchmark: probes cost
README.md - Researcher → Results aggregation: aggregates runs
README.md - Results aggregation → Figures and tables: feeds reports
- Researcher → Explanation pipeline: runs explanations
README.md - Explanation pipeline → Clip dataset: loads video data
src/echoef/xai/pipeline.py - Explanation pipeline → Clip sampling: selects clips
src/echoef/xai/pipeline.py - Explanation pipeline → Attribution methods: computes attributions
src/echoef/xai/pipeline.py - Explanation pipeline → Explanation rendering: renders explanations
- Researcher → Local demo: opens demo
README.md - Local demo → Demo inference: requests inference
app/app.py - Demo inference → EF regressor: loads checkpoints
app/inference.py - Demo inference → Clip dataset: uses clip data
app/inference.py - Demo inference → Attribution methods: explains prediction
app/inference.py - Training sweep → Experiment configuration: loads configuration
src/echoef/train/sweep.py