ZEN-IQA: Zero-Shot Explainable and No-Reference Image Quality Assessment with Vision Language Model (IEEE Access)

Paper

Takamichi Miyata Chiba Institute of Technology

This is the official implementation of ZEN-IQA. The code has been tested on PyTorch 1.13 and CUDA 11.7. To build your environment, run

conda create -n ZEN-IQA python=3.8 -y
conda activate ZEN-IQA
conda install pytorch==1.13.0 torchvision==0.14.0 torchaudio==0.13.0 pytorch-cuda=11.7 -c pytorch -c nvidia
pip install mmcv-full
pip install tqdm ftfy pillow regex einops pandas
git clone https://github.com/mtakamichi/ZEN-IQA
cd ZEN-IQA
pip install -e .

How to run

Test ZEN-IQA on KonIQ-10k dataset

python demo/zeniqa_koniq_demo.py --file_path ..\KonIQ10k\1024x768\ --csv_path ..\KonIQ10k\koniq10k_distributions_sets.csv

Test ZEN-IQA on Live-itW dataset

python demo/zeniqa_liveitw_demo.py --file_path ..\Live-itw\Images\ --csv_path ..\Live-itw\Data

Citation

If our work is useful for your research, please consider citing:

@article{Miyata2024_ZENIQA,
    author = {Takamichi Miyata},
    title = {ZEN-IQA: Zero-Shot Explainable and No-Reference Image Quality Assessment with Vision Language Model},
    journal  = {IEEE Access},
    year = {2024}
}

Acknowledgement

This implementation is heavily based on CLIP-IQA, MMEditing, and CLIP. We thank the original authors for their open-sourcing.