Manas Jyoti Buragohain

I am a software engineer at Magic Leap working on developing object pose estimation solutions for real-world objects. I completed my Master of Science in Robotics from the Robotics Institute at the University of Michigan working with Justin Johnson.

My research focuses on applying deep learning towards 3D reconstruction. I am particularly interested in exploring novel object representation that are feasible for learning-based methods.

Prior to coming to Michigan, I got my undergraduate degree at Delhi Technological University, with a major in Electronics Engineering and a minor in Robotics.

Email | CV | LinkedIn | GitHub

News

[08/2021]    I started as Software Engineer, Perception at Magic Leap.
[04/2021]    I graduated from Univesity of Michigan.
[01/2021]    I am the GSI for EECS 442: Computer Vision with Justin Johnson and David Fouhey.
[08/2019]    I started my Masters program at the Robotics Institute.

Projects

Sparse Neural Generative Inference Based Pose Estimation
Stanley Lewis, Manas Buragohain, Danish Syed, and Bahaa Aldeeb
Course project, EECS 542 Advanced topics in Computer Vision, Fall 2020.
Instructor: David Fouhey.
paper

A particle filter based end-to-end pose estimator where each particle learns latent embedding to infer pose, object likelihood, and re-sampling objective iteratively.

Single Image 3D Reconstruction based on Conditional Generative Adverserial Networks
Danish Syed, Manas Jyoti Buragohain, and and Hansal Shah
Course project, EECS 504 Foundations of Computer Vision, Winter 2020.
Instructor: Andrew Owens.
paper | code

An end-to-end conditional GAN framework for generating 3D objects from single RGB image. We achieve improved qualitative 3D reconstructions as compared to the Pixel2Mesh baseline.

Probabilistic Data Association for Semantic SLAM with Loop Closure Detection
Manas Buragohain, Aohan Mei, Can Jiang, Owen Winship, and Yidong Du
Course project, EECS 568 Mobile Robotics, Winter 2020.
Instructor: Maani Ghaffari.
paper | code

We replicate and improve upon the work of Bowman et. al. with augmentations to object detection framework along with incorporation of loop closure for better offline map generation. I led implementation of the percerption part of the project and it was jointly useful for research.

Publications

Fish species classification using graph embedding discriminant analysis
Snigdhaa Hasija*, Manas Jyoti Buragohain*, and S. Indu
CMVIT 2017
paper

A novel method based on an improved image-set matching approach, which uses Graph-Embedding Discriminant Analysis for fish species classification.

Teaching

EECS 442: Computer Vision (Winter '21)
GSI with Justin Johnson and David Fouhey.

ARCH 660: Visionary Machines - Thesis Development Seminar (Fall '20 )
TA with Matias del Campo


Template