https://github.com/SSL92/hyperIQA

Source code for the CVPR'20 paper "Blindly Assess Image Quality in the Wild Guided by A Self-Adaptive Hyper Network"
https://github.com/SSL92/hyperIQA

Last synced: 27 days ago
JSON representation

Source code for the CVPR'20 paper "Blindly Assess Image Quality in the Wild Guided by A Self-Adaptive Hyper Network"

• Host: GitHub
• URL: https://github.com/SSL92/hyperIQA
• Owner: SSL92
• License: mit
• Created: 2020-03-07T03:17:17.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2023-12-14T09:29:58.000Z (7 months ago)
• Last Synced: 2024-02-23T03:35:11.918Z (4 months ago)
• Language: Python
• Homepage:
• Size: 2.04 MB
• Stars: 299
• Watchers: 8
• Forks: 50
• Open Issues: 34
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Lists

• awesome-image-denoising-state-of-the-art - [Code
• awesome-face-related-list - HyperIQA

README

        
# HyperIQA

This is the source code for the CVPR'20 paper "[Blindly Assess Image Quality in the Wild Guided by A Self-Adaptive Hyper Network](https://openaccess.thecvf.com/content_CVPR_2020/papers/Su_Blindly_Assess_Image_Quality_in_the_Wild_Guided_by_a_CVPR_2020_paper.pdf)".

## Dependencies

- Python 3.6+

- PyTorch 0.4+

- TorchVision

- scipy

(optional for loading specific IQA Datasets)

- csv (KonIQ-10k Dataset)

- openpyxl (BID Dataset)

## Usages

### Testing a single image

Predicting image quality with our model trained on the Koniq-10k Dataset.

To run the demo, please download the pre-trained model at [Google drive](https://drive.google.com/file/d/1OOUmnbvpGea0LIGpIWEbOyxfWx6UCiiE/view?usp=sharing) or [Baidu cloud](https://pan.baidu.com/s/1yY3O8DbfTTtUwXn14Mtr8Q) (password: 1ty8), put it in 'pretrained' folder, then run:

```

python demo.py

```

You will get a quality score ranging from 0-100, and a higher value indicates better image quality.

### Training & Testing on IQA databases

Training and testing our model on the LIVE Challenge Dataset.

```

python train_test_IQA.py

```

Some available options:

* `--dataset`: Training and testing dataset, support datasets: livec | koniq-10k | bid | live | csiq | tid2013.

* `--train_patch_num`: Sampled image patch number per training image.

* `--test_patch_num`: Sampled image patch number per testing image.

* `--batch_size`: Batch size.

When training or testing on CSIQ dataset, please put 'csiq_label.txt' in your own CSIQ folder.

## Citation

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

```

@InProceedings{Su_2020_CVPR,

author = {Su, Shaolin and Yan, Qingsen and Zhu, Yu and Zhang, Cheng and Ge, Xin and Sun, Jinqiu and Zhang, Yanning},

title = {Blindly Assess Image Quality in the Wild Guided by a Self-Adaptive Hyper Network},

booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},

month = {June},

year = {2020}

}

```