How proteus1991/griddehazenet works
GridDehazeNet is a PyTorch image-dehazing project with training and SOTS evaluation workflows for indoor and outdoor images. Testing takes a hazy image and its clear reference from the selected validation dataset, runs the network, measures PSNR/SSIM, and can save dehazed results. Training loads paired crops, optimizes the network with smooth-L1 and perceptual losses, validates, and saves weights. The README identifies test.py as the testing entry point, but it was not excerpted; its internal calls are therefore not asserted.
Main entry points: train.py Training entry · test.py Testing entry · utils.py Metrics and validation · val_data.py Validation pairs
Components 14
Train and evaluate
- Training entry
train.py - Testing entry
test.py
Image datasets
- Training pairs
train_data.py - Train images
data/train - Validation pairs
val_data.py - SOTS images
data/test/SOTS
Dehazing network
- GridDehazeNet
model.py - Residual dense block
residual_dense_block.py
Loss and metrics
- Perceptual loss
perceptual.py - Metrics and validation
utils.py
Model artifacts
- Indoor/outdoor weights
indoor_haze_best_3_6 - Dehazed results
indoor_results - Training logs
training_log
Other components
- Researcher
Connections 18
- Researcher → Training entry: starts training
README.md - Researcher → Testing entry: starts testing
README.md - Training entry → Training pairs: loads pairs
train.py - Training pairs → Train images: reads images
train_data.py - Training entry → Validation pairs: loads validation
train.py - Validation pairs → SOTS images: reads pairs
val_data.py - Training entry → GridDehazeNet: trains
train.py - GridDehazeNet → Residual dense block: uses blocks
model.py - Training entry → Perceptual loss: calculates loss
train.py - Training entry → Metrics and validation: validates
train.py - Metrics and validation → Dehazed results: saves images
utils.py - Metrics and validation → Training logs: writes logs
utils.py - Training entry → Indoor/outdoor weights: loads and saves
train.py - Testing entry → Validation pairs: uses validation data
README.md - Testing entry → GridDehazeNet: runs inference
README.md - Testing entry → Metrics and validation: reports metrics
README.md - Testing entry → Indoor/outdoor weights: loads weights
README.md - Testing entry → Dehazed results: saves images
README.md