Ajay Unagar

Applied Research Scientist

ETH Zurich

About Me

Hi! I am Applied Research Scientist at Geomagical Labs (an IKEA company). I build Computer Vision algorithms for IKEA KREATIV designing app that is being used my millions of customers. My work focuses on building 3D reconstruction pipeline based on indoor RGB scans, which utilizes traditional as well as deep learning based algorithms.

Before Geomagical, I did my MSc in Computational Science from ETH Zurich. I focused on 3D Computer Vision, Reinforcement Learning, and Continual Learning. Before that, I did my undegraduate at at IIT Roorkee in India and worked at ZS Associates as a Data Scientist for 2 years.

Interests

3D Computer Vision
Deep Learning
Reinforcement Learning
Robotics

Education

MSc in Computational Science (Specialization in Robotics), 2019 - 2021

ETH Zurich
BTech in Civil Engineering (Minor in Computer Science), 2013 - 2017

IIT Roorkee, India

News

02/2024	Published a patent on indoor image inpainting.
12/2023	Published a patent on general room layout extraction.
10/2022	Hosted a workshop on Indoor Image Inpainting at ISMAR'22 .
10/2022	We published our work on General Room Layout Estimation at ISMAR'22 in Singapore.
11/2021	Graduated from ETH and started new job as Applied Research Scientist at Geomagical Labs in Palo Alto, CA.
10/2021	Finished my MSc thesis with RSL Lab on learning global path planning for ANYmal robot in a Digital Twin. It was mentioned in a talk at NVIDIA GTC 2021.
06/2021	We published our work "Back to the Feature: Robust Camera Localization" at CVPR'21.
12/2020	Finished a semester project with IBM Research. My project report is online
11/2020	Our work on "Battery calibration using Deep RL" has been accepted at ML4Eng workshop at NeurIPS 2020. The talk is available online.
06/2020	We finished 2nd in "Visual Localization for Autonomous Vehicle" competition organized by VisLocOdomMap workshop at CVPR 2020. Our talk is available online.
09/2019	Started my MSc at ETH Zurich!
07/2017	Joined ZS Associates as a Data Scientist. Working on Machine Learning for Health Care

Publications

Robust Planar Optimization for General 3D Room Layout Estimation

ISMAR'22

Konstantinos-Nektarios Lianos, Luis Puig, Ajaykumar Unagar, Salma Jiddi

Back to the Feature: Learning Robust Camera Localization from Pixels to Pose

CVPR'21

PE Sarlin, Ajay Unagar, M Larsson, H Germain, C Toft, V Larsson, M Pollefeys, V Lepetit, L Hammarstrand, F Kahl, T Sattler

Battery Model Calibration with Deep Reinforcement Learning

NeurIPS'20 Workshop & Energies Journal

Ajay Unagar, Yaun Tian, Manuel Arias Chao, Olga Fink

6-DoF Pose Refinement via Sparse-to-Dense Feature-Metric Optimization

CVPR'20 Workshop

Ajay Unagar, Philipp Lindenberger, Nickolaos Tselepidis, Paul-Edouard Sarlin

Experience

Applied Research Scientist

Geomagical Labs

Nov 2021 – Present Palo Alto

Building and shipping Computer Vision products for millions of IKEA Kreativ customers. Focus on 3D recostruction and deep learning for scene understanding.

Research Assistant

Intelligent Maintenance Systems Lab, ETH Zurich

Jun 2020 – Aug 2021 Zurich

Worked on battery model calibration using reinforcement learning project. Resulted in publication at NeurIPS 2020 workshop and full paper submitted to Energies journal.

Masters Project Intern

IBM Research, Zurich

Jun 2020 – Dec 2020 Zurich

Worked on continual learning for image classification in digital pathology. Developed a method which remember the past knowledge via distillation. This method works in replay-free task-incremental setting.

Machine Learning Engineer

Trexquant Investments

Oct 2017 – Oct 2020 Remote (Part-Time)

Worked on building trading strategies using alpha signals from many sources. Applied techniques for filtering useful trading signals and combining them to make profitable strategies.