Skip to content

Latest commit

 

History

History
56 lines (40 loc) · 1.9 KB

README.md

File metadata and controls

56 lines (40 loc) · 1.9 KB

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".

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 or Baidu cloud (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}
}