Zero-DCE-Res

this is a modify version of Zero-DCE, it use the Residual Block to replace the depthwise separable convolution block.

Requirement

1. python3.8
2. pytorch 2.0
3. torchvision
4. opencv
5. cuda 11.8
6. Pillow 9.5.0
   You can also use conda enviroment to run code.

Floder structure

├── data  
│   ├── test_data # testing data. You can make a new folder for your testing data, like LIME, DICM, and New.  
│   │   ├── DICM   
│   │   └── LIME  
│   │   └── New  
│   └── train_data   
│   │   └── low   
│   │        ├── low.zip # The compressed dataset.   
│   │        └── low.z01  
│   │        └── low.z02  
│   └── result  
│       └── guide.txt  
├── lowlight_test.py # testing code  
├── lowlight_train.py # training code  
├── model.py # Zero-DEC-Res network  
├── dataloader.py  
├── Myloss.py  
├── snapshots  
│   ├── Epoch299.pth #  A pre-trained snapshot (Epoch299.pth)  

Test

Before you run the test, please create the new subfolders in "result" folder which have the same name as the subfolders in "test_data"

python lowlight_test.py

The script will process the image from the subfolders in "test_data" folder, then write them to the subfolders(same name as subfolders in "test_floder") you created in "result"

Train

1. go to the "data/train_data/low" folder
2. unzip the low.zip into the current folder
3. run train script

python lowlight_train.py