Code for the paper: "Troublemaker Learning for Low-Light Image Enhancement"

torch == 1.7.1
torchvision == 0.8.2
timm == 0.6.13

We use python 3.7. Higher version is possible, but ensure that the above environment dependency versions match. For example, a higher version of timm may require a higher version of python.

We provide pretrained Predicting Model (PM) and Enhancing Model (EM) weights (i.e. Pred.pth.tar and Enhance.pth.tar). You can download here.

• Download the pretrained PM and EM model parameters (i.e. Pred.pth.tar and Enhance.pth.tar) and place them in the ckpt folder.

• Put some low-light images in the test folder and run:

python evaluate.py --save_dir ./output/ --resume-Pred ./ckpt/Pred.pth.tar --resume-Enhance ./ckpt/Enhance.pth.tar ./test/

The results will be saved in the output folder.