GamutMLP - A Lightweight MLP for Color Loss Recovery (CVPR 2023)

Hoang M. Le¹, Brian Price², Scott Cohen² and Michael S. Brown¹

¹York University

²Adobe Research

This software is provided for research purposes only and CANNOT be used for commercial purposes.

Project Website: https://gamut-mlp.github.io/

BibTex

Please cite us if you use this code or our dataset:

@InProceedings{Le_2023_CVPR,
    author    = {Le, Hoang M. and Price, Brian and Cohen, Scott and Brown, Michael S.},
    title     = {GamutMLP: A Lightweight MLP for Color Loss Recovery},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2023},
    pages     = {18268-18277}
}

Dataset:

• Test dataset: Download 200 16-bit ProPhoto images
• Train dataset: Download 2000 512x512 16-bit ProPhoto images

Code:

Our source code is for PyTorch platforms. There is no guarantee that the trained models produce EXACTLY the same results, but it should be equivalent.

Setup environment:

• We provide environment.yml for conda, which can be installed with: conda env create -f environment.yml
• NOTE: we only test with Linux system.

Run experiment:

• After downloading the test dataset prophoto_full_16b, you should set the data_root in configs/config.yaml with your own path.
• In configs/dataset/prophoto_full_16b.yaml, dataset_name should be the name of test dataset folder, which is originally prophoto_full_16b.
• How to run the fast MLP:

echo "Run MLP tiny"
python main_run.py experiment=exp_gma_cvpr \
method=mlp_tiny_cudnn \
method.n_neurons=32 \
method.is_trained=False \
method.retrain=True \
method.method_name=mlp_tiny_cudnn32_step10ksam50ogsam5 \
method.sample=50 \
method.og_sample=5 \
method.gpus=3

• Train the meta-init MLP:

python main_run.py experiment=exp_gma_train_meta \
pipeline.meta_inner_steps=10000 \
pipeline.meta_epoch=3

• Run the meta-init fast MLP:

python main_run.py experiment=exp_gma_cvpr \
method=mlp_tiny_cudnn \
method.n_neurons=32 \
method.is_trained=True \
method.retrain=True \
method.method_name=meta_tiny32 \
method.pretrained_model=<replace with project's absolute path here>/pretrained_models/meta_tinycudnn32_metaep3_innersteps10k.pt \
method.n_steps=1200 \
method.sample=50 \
method.og_sample=5 \
method.gpus=0

• Note: see configs for more settings
• Checkout our running scripts for more examples of other baseline methods: scripts/cvpr_2023.sh