Image Adaptation for Colour Vision Deficient Viewers Using Vision Transformers

Datasets

This code runs on the Places365, Oxford Flowers, Abstract Art, WikiArt Abstract, Landscape and Still Life datasets The code expects to find them under data/[places,flowers,abstract_art,wikiart_abstract,wikiart_landscape,wikiart_still_life]

• Dataset download

Dependencies

Dependencies are given in the requirements.txt file. This can be run with

pip install -r requirements.txt

Inside a virtual environment

Submodules

This project has the DINO project as a dependency, with some modifications. This code is not submitted as a git repository for anonymity. Instead, run the following to clone the DINO repository to the directory the code expects:

git clone https://github.com/facebookresearch/dino.git
cd dino
git checkout 7c446df5b9f45747937fb0d72314eb9f7b66930a
git apply ../dino.patch

Running the code

With the above dependencies installed, to prepare data, first run:

python adjoint_filter_dataset.py --dataset_type [dataset_name] --cvd_type [cvd_type] --dataset_split_idx [i]

Where [i] is the dataset index to be processed. By default the dataset is split into 1000 total subsets, so i should range from 0-999, inclusive. The number of total subsets can be changed via command line

To sub-sample the dataset splits according to the scheme described in the paper, run:

python adjoint_sample_datasets.py 

which will create a list of files for the dataset code to read in

Finally, to run the adjoint loop, run:

python adjoint_multires_interp.py  --dataset_type [dataset_name] --cvd_type [cvd_type] --severity [cvd_severity] --dataset_split_idx [i]

To visualise the results, the command

python adjoint_recolour_results.py --dataset_type [dataset_name] --cvd_type [cvd_type] --image_number [i]

will output a grid of the style found in the paper for image [i] of [dataset_name] for mild, moderate, and high severities of CVD type [cvd_type]

Citation

If you use this code, please cite our paper:

@InProceedings{Gillooly_2025_WACV,
    author    = {Gillooly, Thomas and Thomas, Jean-Baptiste and Hardeberg, Jon Yngve and Guarnera, Giuseppe Claudio},
    title     = {Image Adaptation for Colour Vision Deficient Viewers Using Vision Transformers},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    month     = {February},
    year      = {2025},
}