A simplified PyTorch implementation of our GMSNet.

Download the dataset and pretrained model from GoogleDrive.

Extract the files to data folder and save_model folder as follow:

~/
  data/
    SIDD_train/
      ... (scene id)
    SIDD_valid/
      ... (id)
    Syn_train/
      ... (id)
    DND/
      images_srgb/
        ... (mat files)
      ... (mat files)
  save_model/
    gmsnet/ (model name)
      best_model.pth.tar

We provide the code to generate a synthetic dataset using clean images.

The code you can find in utils/syn.

Copy the template code to build your own model:

~/
  model/
    gmsnet.py
    template.py
    ... (your model)

Train your own model:

python train.py --model ... (model name)

Evaluate the trained model (--ensemble for higher score):

python submit_dnd.py --model ... (model name) --ensemble

The results are in result/submit_dnd/bundled

Note that you should upload your results to the DND benchmark website by yourself.