How yvonne-ytwu/deepjscc works
This is a research implementation of Deep JSCC image transmission. Its main workflow loads CIFAR-10 or Kodak images, trains an autoencoder for a chosen compression ratio and channel condition, evaluates reconstructions across SNRs, and combines those results with JPEG/JPEG2000 benchmarks in comparison plots. The graph also preserves the separate robustness-plot workflow and the model, channel, and metric layers. Benchmark internals are not excerpted; those connections rely on README descriptions.
Main entry points: evaluate.py Model Evaluation · train.py Training Loop · main.py Training Entry Point · data_load.py Dataset Loaders
Components 18
Data and Configuration
- Experiment Defaults
config.py - Dataset Loaders
data_load.py
Deep JSCC Codec
- Autoencoder
model.py - Channel Models
utils.py - Image Metrics
utils.py
Experiments
- Training Loop
train.py - Training Entry Point
main.py - Trained Weights
- Model Evaluation
evaluate.py - Codec Benchmarks
benchmark.py
Results and Visualization
- JSON Results
- Comparison Plot Runner
plot_result.py - Robustness Plot Runner
plot_figure4.py - Plotting Functions
visualization.py
Other components
- Researcher
- CIFAR-10
- Kodak Images
- JPEG / JPEG2000
Connections 26
- Researcher → Training Entry Point: starts training
main.py - Training Entry Point → Training Loop: calls training
main.py - Training Loop → Experiment Defaults: reads defaults
train.py - Training Loop → Dataset Loaders: loads datasets
train.py - CIFAR-10 → Dataset Loaders: provides images
data_load.py - Kodak Images → Dataset Loaders: provides images
data_load.py - Training Loop → Autoencoder: trains
train.py - Autoencoder → Channel Models: applies channel
model.py - Training Loop → Image Metrics: computes validation
train.py - Training Entry Point → Trained Weights: saves weights
main.py - Researcher → Model Evaluation: starts evaluation
evaluate.py - Model Evaluation → Dataset Loaders: loads test data
evaluate.py - Model Evaluation → Trained Weights: loads weights
evaluate.py - Model Evaluation → Autoencoder: evaluates
evaluate.py - Model Evaluation → Image Metrics: computes scores
evaluate.py - Model Evaluation → JSON Results: writes metrics
evaluate.py - Researcher → Codec Benchmarks: starts benchmark
README.md - Codec Benchmarks → JPEG / JPEG2000: benchmarks codecs
README.md - Codec Benchmarks → JSON Results: writes benchmark data
README.md - Researcher → Comparison Plot Runner: starts plotting
plot_result.py - Comparison Plot Runner → JSON Results: reads results
plot_result.py - Comparison Plot Runner → Plotting Functions: calls comparison plot
plot_result.py - Researcher → Robustness Plot Runner: starts plotting
plot_figure4.py - Robustness Plot Runner → JSON Results: reads evaluations
plot_figure4.py - Robustness Plot Runner → Plotting Functions: calls robustness plot
plot_figure4.py - Training Entry Point → Plotting Functions: creates training report
main.py