How lijiyidong/segca-net works
SEGCA-Net is a low-light image-enhancement training project. The evidenced workflow loads paired low/high images, prepares training or validation samples, runs HGETNet, computes a task-specific loss, and updates the model; validation scores guide best-checkpoint saving. The README gives only the project name, and the tree omits imported model and utility modules, so their implementations and some metric wiring cannot be verified from the supplied files.
Main entry points: train.py Training loop · data_RGB.py Dataset factories · dataset_RGB.py Image datasets
Components 9
Image data
- Dataset factories
data_RGB.py - Image datasets
dataset_RGB.py
Model training
- Training loop
train.py - HGETNet
- Enhancement loss
Evaluation artifacts
- Validation metrics
- Model checkpoints
train.py - Training logs
train.py
Other components
- Paired images
Connections 8
- Paired images → Dataset factories: provides pairs
- Training loop → Dataset factories: requests datasets
- Dataset factories → Image datasets: constructs
- Training loop → HGETNet: runs forward
- Training loop → Enhancement loss: computes loss
- Training loop → Validation metrics: scores validation
- Training loop → Model checkpoints: saves weights
- Training loop → Training logs: writes scalars