Scale Guided Hypernetwork for Blind Super-Resolution Image Quality Assessment

Paper

TODO

• Code release
• Upload datasets
• clean the code

Introduction

• log: save training log
• nets: define iqa model
• save: save model
• nets: define iqa model
• datasets.py: define datasets
• datasets_deepsrq.py: define datasets for deepsrq
• engine.py: training and test engine
• train_test_IQA.py: setup training and test

Train and Test

First, download datasets used in this paper from here.

Second, change the dataset path in the train_test_IQA.py as follows:

    path = {
        'QADS': 'yourpath/QADS/',
        'CVIU': 'yourpath/CVIU/',
        'Waterloo': 'yourpath/Waterloo/',
    } 

Third, train and test the model using the following command:

python train_test_IQA.py --dataset xxx --netFile xxx --gpuid x --batch_size 64

Some mandatory options:

• --dataset: string, Training and testing dataset, support datasets: 'CVIU' | 'QADS'| 'Waterloo'.
• --netFile: string, IQA model, support models: 'DBCNN' | 'HyperIQA' | 'CNNIQA' | 'Resnet50' | 'JCSAN' | 'DeepSRQ'.
• --gpuid: int, gpu device
• --batch_size: int, Batch size, 64.

Acknowledgement

This project is based on HyperIQA. Thanks for the awesome work.

Citation

Please cite the following paper if you use this repository in your reseach.

@article{fu2023scale,
  title={Scale Guided Hypernetwork for Blind Super-Resolution Image Quality Assessment},
  author={Fu, Jun},
  journal={arXiv preprint arXiv:2306.02398},
  year={2023}
}

Contact

For any questions, feel free to contact: [email protected]