Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/subpic/koniq
KonIQ-10k Deep Learning Models
https://github.com/subpic/koniq
image-quality-assessment
Last synced: 30 days ago
JSON representation
KonIQ-10k Deep Learning Models
- Host: GitHub
- URL: https://github.com/subpic/koniq
- Owner: subpic
- License: mit
- Created: 2019-09-02T11:41:53.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2021-09-29T14:22:37.000Z (over 3 years ago)
- Last Synced: 2023-10-20T23:16:48.980Z (about 1 year ago)
- Topics: image-quality-assessment
- Language: Jupyter Notebook
- Homepage:
- Size: 558 KB
- Stars: 105
- Watchers: 5
- Forks: 22
- Open Issues: 6
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
## NR-IQA models trained on the KonIQ-10k dataset
This is part of the code for the paper ["KonIQ-10k: An ecologically valid database for deep learning of blind image quality assessment"](https://arxiv.org/abs/1910.06180). The included Python 2.7 notebooks rely on the [kutils library](https://github.com/subpic/kutils). The Google colab requires the [ku library](https://github.com/subpic/ku). Project data is available for download from [osf.io](https://osf.io/hcsdy/).
To quickly try out the `Koncept512` model:
```
pip install koncept
```
Please cite the following paper if you use the code or package:
```
@article{koniq10k,
author={V. {Hosu} and H. {Lin} and T. {Sziranyi} and D. {Saupe}},
journal={IEEE Transactions on Image Processing},
title={KonIQ-10k: An Ecologically Valid Database for Deep Learning of Blind Image Quality Assessment},
year={2020},
volume={29},
pages={4041-4056}}
```
## Overview
Google colab notebook, Python 3 compatible:
**[koncept512_train_test_py3_with_kuti.ipynb](https://github.com/subpic/koniq/blob/master/koncept512_train_test_py3_with_kuti.ipynb)** *(updated Sept 2021)*
- Download KonIQ-10k dataset, train the KonCept512 model and test it
- Load a pre-trained KonCept512 model, and use it to predict the quality of an image
Python 2.7 notebooks:
**[train_koncept512.ipynb](https://github.com/subpic/koniq/blob/master/train_koncept512.ipynb), [train_koncept224.ipynb](https://github.com/subpic/koniq/blob/master/train_koncept224.ipynb)**:
- Training and testing code for the KonCept512 and KonCept224 model (on KonIQ-10k).
- Ready-trained model weights for [KonCept512](https://osf.io/uznf8/download) and [KonCept224](https://osf.io/cxtyp/download).
**[train_deeprn.ipynb](https://github.com/subpic/koniq/blob/master/train_deeprn.ipynb)**
- Reimplementation of the [DeepRN](https://www.uni-konstanz.de/mmsp/pubsys/publishedFiles/VaSaSz18.pdf) model trained on KonIQ-10k, following the advice of the original author, Domonkos Varga.
- Re-trained model weights (on SPP features) are available [here](https://osf.io/avyd5/download).
- The features extracted from KonIQ-10k are available [here](https://osf.io/y6brn/download).
**[metadata/koniq10k_distributions_sets.csv](https://github.com/subpic/koniq/blob/master/metadata/koniq10k_distributions_sets.csv)**
- Contains image file names, scores, and train/validation/test split assignment (random).