During this project, I conduct research in the area of deep learning for 3D vision in collaboration with Microsoft Research. I focus on using deep learning techniques for improving multi-sensor fusion performance in mixed reality. The questions that are investigated in this project are:- deep learning for 3D vision- leveraging learned uncertainty for sensor fusion- using deep learning for data reconstruction, noise filtering and outlier removal- implementation of a multi-sensor fusion pipeline on a portable device for mixed reality

As an applied science intern at Amazon, I did a research project in the area of computer vision, where I worked on the following tasks:- adapting and training advanced deep learning models for instance segmentation- exploring and developing different methods to refine synthetic data for training instance segmentation models- implementing and comparing different strategies to train deep neural models for instance segmentation from synthetic data- testing the impact of deep instance segmentation models on the actual robot

During this research project, I develop and implement a novel neural network architecture for robotics scene understanding with the final goal of enabling autonomous systems to compensate for sensor failure (radar, camera, lidar, etc,) during operation and therefore making the system safer. Additionally, I develop a sequential extension of the proposed model using a recurrent neural model. After the training of the implemented model, I test the final model using different widely-used data sets in autonomous driving (e.g. KITTI) and evaluate its performance using various baselines.