Machine Perception
Machine Perception is a lecture taught at ETH Zurich, tackling the most recent deep learning approaches and architectures for perception tasks, such as Variational Auto-Encoders, Normalizing Flows or Generative Adversarial Networks. Towards the end of the semester, each student had the opportunity to get some hands-on experience by implementing a state-of-the-art network for a concrete application, in a team of 3 people. My group focused on estimating human Optical Flow by building our own architecture, Seg-Net, on top of Nvidia’s PWC-Net, to achieve better results.
A detailed description on our work and evaluated results are available in the below report.
Report
