How enrico-favale/deep-image-restoration works
This is a research codebase for restoring Fourier- and wavelet-degraded BSD500 images with convolutional autoencoders and Restormer. Its principal workflow is to create compressed images, pair them with originals, train a restoration model, run inference, and compare restoration quality with image metrics. The graph also shows training metadata and visualization as supporting capabilities. Some relationships are documented only at a high level in the README; unsampled notebook wiring and implementation details are not asserted.
Main entry points: trainer.py Model trainer · fourier.py Fourier compression · wavelet.py Wavelet compression · inference.py Image restoration
Components 17
Image compression
- Fourier compression
src/compression/fourier.py - Wavelet compression
src/compression/wavelet.py - Compressor interface
src/compression/compressor.py - Image file utilities
src/compression/utils.py
Restoration learning
- Image-pair dataset
src/utils/preprocessing.py - Model trainer
src/utils/trainer.py - Convolutional autoencoder
src/models/autoencoder.py - Restormer
src/models/restormer.py
Inference and assessment
- Image restoration
src/utils/inference.py - Quality evaluation
src/utils/evaluation.py - Restoration plots
src/utils/plot.py
Training support
- Experiment logger
src/utils/logger.py
Other components
- Researcher
- Original images
- Compressed images
- Model checkpoints
- Restored image
Connections 26
- Researcher → Fourier compression: compresses with
README.md - Researcher → Wavelet compression: compresses with
README.md - Original images → Fourier compression: provides images
src/compression/fourier.py - Original images → Wavelet compression: provides images
src/compression/wavelet.py - Fourier compression → Compressor interface: implements
src/compression/fourier.py - Wavelet compression → Compressor interface: implements
src/compression/wavelet.py - Fourier compression → Image file utilities: loads and saves
src/compression/fourier.py - Wavelet compression → Image file utilities: loads and saves
src/compression/wavelet.py - Fourier compression → Compressed images: produces
src/compression/fourier.py - Wavelet compression → Compressed images: produces
src/compression/wavelet.py - Compressed images → Image-pair dataset: pairs with originals
src/utils/preprocessing.py - Original images → Image-pair dataset: provides targets
src/utils/preprocessing.py - Researcher → Model trainer: trains model
README.md - Model trainer → Image-pair dataset: loads and splits
src/utils/trainer.py - Model trainer → Convolutional autoencoder: trains supplied model
src/utils/trainer.py - Model trainer → Experiment logger: records experiment
src/utils/trainer.py - Convolutional autoencoder → Experiment logger: logs architecture
src/models/autoencoder.py - Model trainer → Model checkpoints: saves best model
- Researcher → Image restoration: restores image
README.md - Model checkpoints → Image restoration: provides weights
src/utils/inference.py - Image restoration → Convolutional autoencoder: runs supplied model
src/utils/inference.py - Image restoration → Restormer: runs supplied model
- Image restoration → Restored image: returns image
src/utils/inference.py - Researcher → Quality evaluation: evaluates quality
README.md - Quality evaluation → Restored image: measures restoration
- Researcher → Restoration plots: compares results