JGyoung33 / hdrnet-pytorch

Unofficial PyTorch implementation of 'Deep Bilateral Learning for Real-Time Image Enhancement', SIGGRAPH 2017 https://groups.csail.mit.edu/graphics/hdrnet/

Deep Bilateral Learning for Real-Time Image Enhancements

Unofficial PyTorch implementation of 'Deep Bilateral Learning for Real-Time Image Enhancement', SIGGRAPH 2017 https://groups.csail.mit.edu/graphics/hdrnet/

Python 3.6

Dependencies

To install the Python dependencies, run:

pip install -r requirements.txt

Datasets

Adobe FiveK - https://data.csail.mit.edu/graphics/fivek/

Usage

To train a model, run the following command:

python train.py --test-image=./DSC_1177.jpg --dataset=/dataset_path --lr=0.001

To get all train params run:

python train.py -h

To test image run:

python test.py --checkpoint=./ch/ckpt_0_4000.pth --input=./DSC_1177.jpg --output=out.png

Known issues

• Torch F.grid_smaple doesn't have triliniear interpolation that was used in original network(which is strange cause it can use 3D image as input), that's make things worse. Hope they will fix this, until that will try fix this somehow.
• Only PointwiseNN implemented currently
• Dataset has no augmentation which making training difficult
• No raw HDR input