Shrisudhan Govindarajan

I am a Data & Applied Scientist at Microsoft India (R&D), Hyderabad. I completed my Dual Degree undegraduate from IIT Madras with major 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 and Shubham Dhage from Qualcomm, India as a part of Qualcomm Innovation Fellowship, 2021-22.

My main research interest lies at the intersection of Computer vision and Computational Photography. I am recently drawn towards the latest research works in NeRF and Diffusion, and their intersection for 3D scene generation. I am currently looking for PhD opportunities to work on Computer Vision and Computational Phototgraphy, especially Generative models, Neural Rendering and Implicit Neural Representations.

Link to my Masters Thesis              |              Link to my Bachelors Thesis

Email  /  CV  /  Google Scholar  /  Twitter  /  Github

profile photo
Research

I am interested in solving problems at the intersection of compter vision and computational photography. Much of my research till now is on computer vision techniqiues used for addressing computational photography related problems such as Light Field Imaging, Underwater Imaging, HDR Imaging, etc.

Synthesizing Light Field Video from Monocular Video
Shrisudhan Govindarajan, Prasan Shedligeri, Kaushik Mitra
ECCV, 2022 [oral]
project / pdf / supp / arXiv / code / cite

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

Battery Prognostics: Estimation of Remaining Operational Time of batteries using convolution and temporal-correlation
Shrisudhan Govindarajan, Deep Singh, Arunachalam N,
Bachelors Thesis
pdf

In this study, we propose a novel deep neural network (DNN) based architecture which uses temporal information to learn the changes in the battery across time and utilizes that in-coherence with the available parameters for estimating accurate remaining operational time.

Caching in DNNs - Speeding up Inference for similar inputs
Shrisudhan Govindarajan, Pratyush Kumar
pdf / code

We propose to use caching in DNNs to improve the inference speed for classification problems and also improve the robustness of the network towards brightness, contrast variations, and increase immunity towards adversarial attacks.

Invited Talks

Mobile Intelligent Photography and Imaging (MIPI) workshop, ECCV 2022
Invited Talk: Synthesizing Light Field Video from Smartphones
ECCV, 2022
workshop / slides / youtube

In the last 2 decades, we have seen a revolution in mobile imaging with improvements in both the hardware and software. However, these cameras capture only a 2D projection of our rich 3D world. In this talk we propose a self-supervised learning technique to reconstruct light field(containing 3D information) video from simple smartphone camera configurations, namely monocualr camera and stereo camera(2D projections). We propose various novel techniques to address the challenges associated with these camera configurations in our attempt to synthesize structurally and temporally consistent light field video.

Vision India, ICVGIP 2022
Invited Talk: Synthesizing Light Field Video from Monocular Video
ICVGIP, 2022
conference / slides

Learning-based techniques which solve the ill-posed problem of LF reconstruction from sparse (1, 2 or 4) views have significantly reduced the requirement for complex hardware. LF video reconstruction from sparse views poses a special challenge as acquiring ground-truth for training these models is hard. In this talk we propose a self-supervised learning-based algorithm for LF video reconstruction from monocular videos. We propose novel techniques to address the limitations of monocular input sequences for light field synthesis task, like difficulty in occlusion handling and depth scale perception.

Professional Experience

Microsoft R&D, India - Search Technology Center India
Data and Applied Scientist
July, 2022 - Present

In the current version of Microsoft Teams, Office and Sharepoint space, for a given search, we see multiple entity sets, like People suggestions, Message suggestions, File suggestions, Calendar suggestions and others. I work on developing a ranking algorithm to rank these different entity sets based on their relevance to the searched query and previous user interaction.

Microsoft R&D, India - Search Technology Center India
Data and Applied Scientist Intern
May, 2021 - July, 2021

Developed an ranking algorithm to rank related suggestions for a query based on the relatedness and usefulness of the suggestion in an Enterprise-level(Microsoft Bing Work vertical) setup.

AutoInfer Pvt. Ltd.
Deep Learning Intern
June, 2020 - August, 2020

Developed a Generative Network inspired by the Layout2Image algorithm to generate realistic documents from user-specified layouts. Built a table detection network inspired by LayoutLM algorithm which extracts textual and image features from the document to detect tables and information.

Education
Indian Institute of Technology Madras
Integrated Dual Degree in Data Science and Mechanical Engineering(Honours)
July '17 - May '22
Miscellaneous

Participated in Intern IIT 2018, held at IIT Bombay. We presented a prototype of driver assistance system with lane detection, object detection and, sign and signal detection features(Link).

Some of the best experiences I've had in my undergraduate life is due to Computational Imaging Group( Link). The seniors and people there are some of the best and loveliest you can find in IIT Madras. I am highly indebted for being a part of the group.

I have also had the pleasure to be a part of the IIT Madras Wolves(football/soccer team).

In my free time, I love to watch Movies, TV shows and you can most definitely find me listening to music at any time of day.



This template is stolen from here.