Deep Learning-based Image Compression

Status: available
Supervisors: Manuel Keglevic, Matthias Wödlinger


so smart

Traditional lossy image compression techniques like JPEG are pretty old and are honestly only used today because they are built-in in every device that can process images from your camera to your smart watch (and definitely your smart toaster). Since Deep Learning-based end-to-end solutions work pretty well in other Computer Vision fields, of course somebody had the idea to apply this to image compression as well and here we are now. If you want to see some insane results head over to the HiFiC Interactive Demo and be amazed how bad JPEG actually is.

In contrast to other Deep Learning fields though, Deep Learning-based image compression is more about clever models and losses and less about deeper and deeper models and augmenting your data the right way to get what you want. So if you always wanted to understand how this Deep Learning stuff actually works this is a great field to do research in.


The goal of this work is to develop an end-to-end image compression pipeline adapted to a specific computer vision application, e.g. medical images, satellite images, or any other field where there are public image datasets available. As a starting point, we will provide you with a baseline image compression model similar to [1], point you to literature you need to read like [2] and of course help you at every step of your way. In the end you should have a working image compression pipeline that works better than anything else for the type of data you’re concerned with – which you will show by evaluating your model.


  • Literature Review – getting to know the methods
  • Implementation & tinkering
  • Evaluation
  • Written Report/Thesis and final presentation

Helpful experience

  • Python and numpy
  • Computer Vision applications and frameworks
  • Machine Learning frameworks like PyTorch or Tensorflow


[1] David Minnen, Johannes Ballé, George Toderici, “Joint Autoregressive and Hierarchical Priors for Learned Image Compression“, NIPS 2018

[2] Yueyu Hu, Wenhan Yang, Zhan Ma, Jiaying Liu, “Learning End-to-End Lossy Image Compression: A Benchmark“, Preprint 2020