iqa-contentsep's Introduction

No Reference Opinion Unaware Quality Assessment of Authentically Distorted Images

Nithin C. Babu, Vignesh Kannan and Rajiv Soundararajan

Official pytorch implementation of the WACV'23 paper:
No Reference Opinion Unaware Quality Assessment of Authentically Distorted Images

Performance evaluation

Pre-trained weights

Google drive link for pre-trained weights:

Copy the files to ./pre_trained_models/ folder.

Setting up pristine patches

Google drive link for pre-selected pristine patches link. Copy the downloaded file to the dataset_images folder as ./dataset_images/pristine_patches_096_0.75_0.80.hdf5 .

Testing code

Sample testing code for evaluating the final model on different authentically distorted datasets.

python ./evaluate_model.py --dataset LIVEC --model_weights ./pre_trained_models/auth_ft_cd.pth --eval_result_dir ./results/auth_ft_cd/
python ./evaluate_model.py --dataset KONIQ --model_weights ./pre_trained_models/auth_ft_cd.pth --eval_result_dir ./results/auth_ft_cd/
python ./evaluate_model.py --dataset LIVEFB --model_weights ./pre_trained_models/auth_ft_cd.pth --eval_result_dir ./results/auth_ft_cd/
python ./evaluate_model.py --dataset CID --model_weights ./pre_trained_models/auth_ft_cd.pth --eval_result_dir ./results/auth_ft_cd/

Citation

If you find this work useful for your research, please cite our paper:

@InProceedings{iqa_content_sep,
    author    = {Babu, Nithin C. and Kannan, Vignesh and Soundararajan, Rajiv},
    title     = {No Reference Opinion Unaware Quality Assessment of Authentically Distorted Images},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    month     = {January},
    year      = {2023},
    pages     = {2459-2468}
}

Recommend Projects