Autoencoders & VAEs

Convolutional Autoencoders have an extensive and successful record in applications such as representation learning, dimensionality reduction and learning data transformations. Recently, the concept of autoencoding has received even more attention with a rise of generative applications, which also make use of this extremely versatile concept. In this workshop, we will first introduce the concept and inner workings of autoencoders. We will then have a hands-on session with attendees implementing and training autoencoders on their own. The application focus will be chosen from within the following domain(s):

  • Image completion: realistically fill in missing or corrupted parts of an image
  • Image colorization: turn grayscale/black & white images into realistic color images
  • Image style transfer: transform images into different styles, resembling certain painters or photographers
  • Image denoising: learn to suppress typical image noise, producing a clean image
  • Image segmentation: generate image masks that highlight object types and their extent in an image

Even though we will focus on just one of the above applications, attendees will acquire all skills necessary to build their own models solving others of the above tasks.

Upon completion of this workshop, attendees will be able to:

  • Understand the concept of autoencoders and how they work
  • Design basic network architectures for autoencoders by themselves
  • Successfully train models for solving the above, or similar computer vision tasks
  • Expand their knowledge into more advanced, generative versions of this principle, such as variational autoencoders (VAE) after attendance.

Open in Full Screen