Blind Image Quality Assessment via Transformer Predicted Error Map and Perceptual Quality Token

Environment

$ pip install -r requirements.txt

$ conda env create -f environment.yaml

Datasets

In this work we use 6 datasets (LIVE, CSIQ, TID2013, KADID10K, LIVE challenge, KonIQ)

Training

1. Pre-train model for EM.

   $ python train_pre.py

2. Final model for score prediction.

   $ python train_final.py

Pretrained Models

Pretrained models will be released soon.

Visualization

1. Predicted Error Maps

2. Perceptual Attention Maps