This is the repository contains the official pytorch implementation of the paper Toward a blind image quality evaluator in the wild by learning beyond human opinion scores, Zhihua Wang, Zhiri Tang, Jianguo Zhang, and Yuming Fang, Pattern Recognition, 2023.
You can download the pre-trained weights adapting to KonIQ-10k and SPAQ.
- You need to download Waterloo Exploration Database (https://ece.uwaterloo.ca/~k29ma/exploration/) first, and then leverage the distortion generation codes to simulate distorted images.
- The FR-IQA models for pseudo-label predictions includes FSIMc, SR-SIM, NLPD, VSI, MDSI and GMSD, released by respectively authors.
- Randomly sample image pairs and assign binary pseudo-labels.
You can run the Main.py for training and the test_SPAQ.py and test_KonIQ.py for testing
If you find the repository helpful in your resarch, please cite the following papers.
@article{wang2023toward,
title = "Toward a blind image quality evaluator in the wild by learning beyond human opinion scores",
author = "Zhihua Wang and Zhi-Ri Tang and Jianguo Zhang and Yuming Fang",
year = "2023",
volume = "137",
journal = "Pattern Recognition"}