Varun Sundar

I am a first second third fourthfifth year PhD student at the University of Wisconsin Madison, working with Mohit Gupta at the WISION Lab.

Previously, I was an undergraduate at the Indian Institute of Technology Madras, where I obtained a bachelors degree in Electrical Engineering. At IIT Madras, I was associated with the Computational Imaging Lab, headed by Kaushik Mitra.

I spent the summer of 2019 working with Ram Nevatia at the University of Southern California. I've also been fortunate to be work with some amazing mentors, including Salaman Siddique and Arka Sadhu.

Email  /  CV  /  Google Scholar  /  Twitter  /  Github

I'm interested in computer vision, computational imaging and machine learning. Broadly, I wish to harness a conjuction of newer hardware and smarter algorithms to tackle a variety of problems. My undergraduate research focussed on developing deep learning techniques to recover high-quality images from lensless cameras.

• (June 19th, 25) One first-authored paper accepted to ICCV! See you in Hawaii!
• (June 3rd, 25) One co-first-authored paper accepted to ICCP! See you in Toronto!
• (May 19th, 25) I will be interning at Snap, New York City this summer!
• (Jan 10th, 25) XRDS magazine published our invited article!
• (Aug 1st, 24) Our single-photon demo was recognized with the "best in show" award at SIGGRAPH E-Tech!
• (Jul 19th, 24) Demoing single-photon applications at SIGGRAPH. Say hi if you're at Denver next week!
• (Feb 26th, 24) One first-authored paper accepted at CVPR! See you in Seattle!
• (Jul 13th, 23) One first-authored paper accepted at ICCV as an oral presentation!

A learnable temporal aggregation layer that transforms many image- and video-based neural networks into neural networks that work on the acquisition of quanta image sensors.

Reducing avalanche power consumption in passive video capture with SPAD sensors, by restricting measurements to salient pixels identified dynamically from the data.

An invited article in the Crossroads (XRDS) magazine, explaining single-photon imaging in an accessible manner.

A demonstration of single-photon imaging capabilities---including high-speed, high-dynamic range, and low-light vision---at SIGGRAPH's emerging technologies track.

Single-photon cameras can not only emulate event cameras, but also facilitate a generalized space of intensity-preserving event cameras.

Photon-level software-defined cameras can emulate a diverse range of imaging modalities.

Color filter and algorithm design for single-photon color imaging in low light.

Single-Photon Avalanche Diodes (SPADs) can be operated as extremely high-speed binary imagers and find a synergistic application in Structuered Light.

Reproducibility Report on the ICML 2020 paper "Rigging the Lottery: Making All Tickets Winners".

Deep Atrous Guided Filter for Image Restoration in Under Display Cameras
Varun Sundar*, Sumanth Hegde*, Divya K Raman, Kaushik Mitra.
ECCV RLQ-TOD Workshop, 2020
* denotes equal contribution.
project page / arXiv / video / bibtex

Guided Filters when incorporated in a deep network can efficiently recover severely degraded, mega-pixel resolution images.

FlatNet: Towards Photorealistic Scene Reconstruction from Lensless Measurements
Salman S. Khan*, Varun Sundar*, Vivek Boominathan, Ashok Veeraraghavan, Kaushik Mitra.
IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020
* denotes equal contribution.
project page / arXiv / video / bibtex

We propose a general learning based framework to recover photorealistic scenes from lensless captures for both separable and non-separable forward models.