README.md

This is the official repository for the implementation of the paper: "Enhancing image quality prediction with self-supervised visual masking". You can find the paper here: https://arxiv.org/abs/2305.19858

The list of the enhanced metrics:

• L1
• L2
• PSNR
• SSIM
• VGG
• LPIPS
• DISTS

The code for the enhanced metric for MAE can be run simply with:

python Masked_L1.py --ref images/ref.BMP --dist images/dist.BMP

The other metrics can also be run in the same way.

The code was tested under Debian GNU/Linux 11.

Dependencies:

pytorch-cuda==11.7

numpy==1.23.3

torchvision==0.14.0

pillow==9.2.0

For the citation:

@misc{çoğalan2024enhancing,
      title={Enhancing image quality prediction with self-supervised visual masking}, 
      author={Uğur Çoğalan and Mojtaba Bemana and Hans-Peter Seidel and Karol Myszkowski},
      year={2024},
      eprint={2305.19858},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}