AI

Machine Learning in Linux: FBCNN – JPEG artifacts removal

In Operation

The project’s repository provides 4 models:

  • Grayscale JPEG images – main_test_fbcnn_gray.py
  • Grayscale JPEG images trained with double JPEG degradation model – main_test_fbcnn_gray_doublejpeg.py
  • Color JPEG images – main_test_fbcnn_color.py
  • Real-world JPEG images – main_test_fbcnn_color_real.py

The project provides test sets to be used for the 4 models which are stored in the testsets directory. When you run a script (e.g. python main_test_fbcnn_color_real.py) it automatically downloads the relevant mode, runs through the images in the relevant testsets directory and output the results to the test_results directory.

To test your own JPEGs, copy them to the relevant subdirectory of the testsets directory.

Each script contains a quality factor list. By setting different quality factors, the trade-off between artifacts removal and details preservation is controlled.

Here’s an example JPEG suffering from artifacts.

FBCNN
Click image for full size

And the output with different quality factors:

QF=10

FBCNN
Click image for full size

When you use low QF numbers, most artifacts together with some texture details are removed.

QF=50

FBCNN
Click image for full size

QF=90

FBCNN
Click image for full size

Summary

FBCNN is an interesting project. It offers flexible models to obtain desirable results with fewer artifacts.

There is training code available.

Website: github.com/jiaxi-jiang/FBCNN
Support:
Developer: Jiaxi Jiang, Kai Zhang, Radu Timofte
License: Apache License 2.0

FBCNN is written in Python. Learn Python with our recommended free books and free tutorials.

Artificial intelligence icon For other useful open source apps that use machine learning/deep learning, we’ve compiled this roundup.

Pages in this article:
Page 1 – Introduction and Installation
Page 2 – In Operation and Summary

Subscribe
Notify of
guest

This site uses Akismet to reduce spam. Please read our FAQ before making a comment.

0 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments