fzp0424/MonotonicIQA

This project contains all the codes used in the Monotonic IQA model.

MonotonicIQA

The code for [2209.10451] Learning from Mixed Datasets: A Monotonic Image Quality Assessment Model (arxiv.org)

Requirements:

Python 3+ PyTorch 1.4+ Matlab Successfully tested on Ubuntu 20.04

Usage

Sampling images from each datasets(Matlab)

sample_name.m

Mixing all the sampled images (Matlab)

combine_pmtrain.m

Train on the mixed datasets for 10 sessions

python Main.py --train True --network basecnn --representation BCNN --batch_size 32 --image_size 384 --lr 3e-4 --decay_interval 3 --decay_ratio 0.9 --max_epochs 24 --backbone resnet34

Get scores

python Main.py --train False --get_scores True

Result analysis

Compute weighted PLCC/SRCC: calculate_mean.m

Acknowledgement

zwx8981/UNIQUE: The repository for 'Uncertainty-aware blind image quality assessment in the laboratory and wild' and 'Learning to blindly assess image quality in the laboratory and wild' (github.com)