Robust Semantic Segmentation
The work was done in collaboration at Prof. Wolfram Burgard’s Autonomous Intelligent Systems, University of Freiburg with Abhinav Valada.
The project involves designing an innovative learnable mechanism to perform robust semantic segmentation in adverse conditions. The model learns how to dynamically weigh different modalities and trust them based on their quality resulting in improved segmentation. We also perform experiments on an autonomous car robot, Viona to drive in forested terrain based only on the perception input of the proposed CNN model.
Code and Data
Videos and Demos