Shrisudhan Govindarajan

I am currently a Ph.D. student in Computing Science, Simon Fraser University(SFU), advised by Prof. Andrea Tagliasacchi. Before this, I was a Data & Applied Scientist at Microsoft India (R&D), Hyderabad. I completed my Dual Degree from IIT Madras with Masters in Data Science.

I've had the pleasure of working with Prof. Kaushik Mitra from IIT Madras, on self-supervised light field synthesis. I've also had the chance to worked with Pawan Baheti from Qualcomm, India as a part of Qualcomm Innovation Fellowship, 2021-22.

My main research interest lies at the intersection of Computer vision and Computer Graphics. I am recently drawn towards the latest research works in NeRF and 3DGS, and their intersection for 3D scene representation. I am mainly interested in working towards developing generalizable 3D scene representations.

Email  /  CV  /  Google Scholar  /  Twitter  /  Github

A representation for neural fields combining the characteristics of fast training NeRF methods that rely on Eulerian grids (i.e.~InstantNGP), with those that employ points equipped with features as a way to represent information (e.g. 3D Gaussian Splatting or PointNeRF).

We show that via a simple modification, one can obtain neural fields that are low-pass filtered, and in turn show how this can be exploited to obtain a frequency decomposition of the entire signal.

We collect the first 3-view dual-pixel video dataset, dpMV, and show that dual-pixel methods for light-field reconstruction outperform purely monocular methods, especially in challenging foreground-background separation regions using faithful guidance from dual pixels.

We propose a self-supervised learning technique to reconstruct light field from monocular video with following novelties: an adaptive low-rank representation for each scene, an explicit disocclusion handling technique, and a novel supervised refinement block(optional) that exploits available ground truth Light Field image dataset.

This template is stolen from here.