Home

Awesome

IQT

Unofficial implementations of CVPR2021 paper "Perceptual Image Quality Assessment with Transformers" (paper link: https://openaccess.thecvf.com/content/CVPR2021W/NTIRE/papers/Cheon_Perceptual_Image_Quality_Assessment_With_Transformers_CVPRW_2021_paper.pdf)

This method accomplishes 1st in the NTIRE2021 Perceptual Image Quality Assessment (PIQA) challenge.

The environmental settings are described below. (I cannot gaurantee if it works on other environments)

Train & Validation

First, you need to download weights of InceptionResNetV2 pretrained on ImageNet database.

Second, you need to download the PIPAL database.

After those settings, you can run the train & validation code by running "train.py"

Belows is the performance on PIPAL database

Inference

First, you need to specify variables in "test.py"

After those settings, you can run the inference code by running "test.py"

Acknolwdgements

We refer to the following website to implement the transformer (https://paul-hyun.github.io/transformer-01/)