SMBU-MM / OF_BIQA

Learning from multiple annotators for blind image quality assessment in the wild

Opinin_Free_BIQA

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.

The generation of pseudo-labeled dataset:

1. 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.
2. The FR-IQA models for pseudo-label predictions includes FSIMc, SR-SIM, NLPD, VSI, MDSI and GMSD, released by respectively authors.
3. Randomly sample image pairs and assign binary pseudo-labels.

Train & Test

You can run the Main.py for training and the test_SPAQ.py and test_KonIQ.py for testing

Citation

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"}