Jan Czarnowski

AI/SLAM Researcher

Imperial College London

Biography

Hi! My name is Jan Czarnowski, I am a PhD Student at Imperial College London under the supervision of Andrew Davison. Prior to that I led a team of robotics researchers in PIAP, Warsaw. My research focuses on integrating Deep Learning with Dense Visual SLAM in order to push the boundaries of spatial perception with a single camera. I enjoy building systems that work in real-time and produce good-looking 3D visualisations and probably spend too much time on computer games.

Interests

Visual SLAM
Deep Learning
Reinforcement Learning

Education

PhD in Computing Research, 2019

Imperial College London
MEng in Robotics, 2015

Warsaw University of Technology
BSc in Robotics, 2013

Warsaw University of Technology

Research Highlights

Full list of publications available on Google Scholar

Michael Bloesch, Jan Czarnowski, Ronald Clark, Stefan Leutenegger, Andrew J. Davison

April 2018 CVPR 2018

CodeSLAM - Learning a Compact, Optimisable Representation for Dense Visual SLAM

Joint work with Michael Bloesch on learning a compact representation for depth maps. We showed that it is possible to use the latent representation learned by a modified auto-encoder network as an alternative manifold for optimization. We have demonstrated a simple SLAM system using the proposed method that is able to reconstruct scenes where conventional methods fail.

This work has received the Best Paper Honorable Mention award at CVPR 2018.

PDF Video

Jan Czarnowski, Stefan Leutenegger, Andrew J. Davison

August 2017 ICCV 2017 Workshop

Semantic Texture for Robust Dense Tracking

This work demonstrates that it is possible to perform dense whole-image alignment on feature activations coming from an off-the-shelf CNN classifier. These features form a pyramid of varying semantic levels which, optimized in a coarse-to-fine manner, allows for much more robust camera tracking.

PDF Video

Ronald Clark, Michael Bloesch, Jan Czarnowski, Stefan Leutenegger, Andrew J. Davison

September 2016 ECCV 2018

LS-Net: Learning to Solve Nonlinear Least Squares for Monocular Stereo

Sum-of-squares objective functions are very popular in computer vision algorithms. However, these objective functions are not always easy to optimize. The underlying assumptions made by solvers are often not satisfied and many problems are inherently ill-posed. In this paper, we propose LS-Net, a neural nonlinear least squares optimization algorithm which learns to effectively optimize these cost functions even in the presence of adversities.

PDF Video