/koniq-PyTorch

A Pytorch implementation of No-Reference Image Quality Assessment (NR-IQA) models trained on the KonIQ-10k dataset

Primary LanguageJupyter NotebookMIT LicenseMIT

koniq-PyTorch

A Pytorch reproduction of No-Reference Image Quality Assessment (NR-IQA) models trained on the KonIQ-10k dataset, proposed in the paper "KonIQ-10k: An ecologically valid database for deep learning of blind image quality assessment".

The code is based on koncept512_train_test_py3.ipynb provided in the official repository.

Download the KonIQ-10k dataset with ground truth:

wget "http://datasets.vqa.mmsp-kn.de/archives/koniq10k_512x384.zip"

wget "https://github.com/subpic/koniq/blob/master/metadata/koniq10k_distributions_sets.csv"

unzip koniq10k_512x384.zip

Train/test the optimal koncept512 model in the paper:

-Traing/test code in koncept512_train_test_pytorch.ipynb.

-The pre-trained model is also available.