Low light image enhancement using MIRNet in PyTorch.
Check out my blog for thorough understanding.
This model achieves 22.97 PSNR on LoL dataset's evaluation set.
Input image:point_right: Restored/Predicted Image :point_right: Ground Truth image
- This model uses LoL dataset.
- All the images were resized to 64x64 during training.
- It uses 3 Residual Recurrent Groups(RRGs) where every RRG contains 2 Multi-scale residual blocks(MSRBs).
- You can see more results in
Resultsdirectory.