hit counter
Viraj Shah
Viraj Shah

Viraj Shah

PhD Student, Electrical Engineering,
University of Illinois, Urbana-Champaign


I am Viraj, currently a PhD student in Electrical Engineering at University of Illinois, Urbana-Champaign. I work at Computer Vision and Robotics Lab. Since May 2021, I an Applied Science intern at Amazon with Dr. Aleix Martinez.

My research interests lie at the intersection of Computer Vision and Imaging Inverse Problems. I work on developing algorithms to leverage novel Generative Models and Deep Learning techniques in solving imaging and image manipulation tasks.

Previously, I completed B.Tech. in EE from IIT Roorkee in 2016 and joined PhD program at Iowa State University where I earned MS in ECE before transferring to UIUC. At DICE Lab, ISU, I worked under the guidance of Dr. Chinmay Hegde.

I explored Inverse problems in Imaging (e.g. compressive sensing) and Generative Models in past through my research projects. Please check out the Projects and Publications section, or my resume to learn more!

For Summer 2019, I was research intern at Siemens Healthineers in Dr. Mariappan Nadar's group working on ML-based MR Reconstuction problems.

During my undergrad, I got an opportunity to work at Indian Space Research Organization (ISRO)(2013); Robot Soccer Lab, National Uiversity of Malaysia (2014); and Telecom SudParis.

For B. Tech. project, I worked under guidance of Dr. Gopinatha Pillai on Answer sentence selection.

Projects & Publications

project name

Solving Inverse Problems using GAN priors

We leveraged the ability of Generative Adversarial Networks (GAN) of learning the real data distribution by using the Generator function as a prior on natural images. We developed an algorithm with provable guarantees which which projects the estimated value in the space of the Generator function to obtain state-of-the-art results through Gradient Descent Optimization on variety of inverse problems such as de-noising and compressed sensing.

Paper accepted for International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2018.

Extended Journal version is under review at IEEE TSP.

Extended Journal version (pdf)

ICASSP 2018 version (pdf)

Code (on github)

project name

Generative Priors for Solving Compressive Phase Retrieval

In this work, we propose replacing the sparsity/support priors with generative priors and propose two algorithms to solve the phase retrieval problem. Our proposed algorithms combine the ideas from AltMin approach for non-convex sparse phase retrieval and projected gradient descent approach for solving linear inverse problems using generative priors.

Paper accepted for International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2019.

ICASSP 2019 version (pdf)

project name

Physics-aware Generative Modeling

The challenge is to synthesize microstructures, given a finite number of microstructure images, and/or some physical invariances that the microstructure exhibits. Our Generative Invarient Network (GIN) model explicitly enforces known physical invariances by replacing the traditional discriminator in a GAN with an invariance checker.

Paper accepted for Neural Information Processing Systems 2018 Workshop on Machine Learning for Molecules and Materials.

Paper accepted for AAAI Conference on Artificial Intelligence, 2019.

arXiv version (pdf)

AAAI 2019 version (pdf)

NeurIPS 2018 workshop version (pdf)

Poster (pdf)

Our submission to MRS Open Data Challenge (pdf)

project name

Signal Reconstruction from Modulo Observations

We consider the problem of reconstructing a signal from under-determined modulo observations. Signal reconstruction under this model is a challenging ill-posed problem. We propose a novel approach to (rigorously) solving the inverse problem, inspired by recent advances in algorithms for phase retrieval under sparsity constraints. We prove that given a sufficient number of measurements, our algorithm perfectly recovers the underlying signal.

Paper accepted for IEEE Global Conference on Signal and Information Processing (GlobalSIP), 2019.

Extended version accepted for publication in EURASIP Journal on Advances in Signal Processing, 2021.

Find out more in extended version of our paper on arXiv (pdf)

project name

Reconstruction from Periodic Nonlinearities for HDR imaging

For the problem of reconstructing signals and images from periodic nonlinearities, we design a measurement scheme that supports efficient reconstruction; moreover, our method can be adapted to extend to compressive sensing-based signal and image acquisition systems. Our techniques can be potentially useful for reducing the measurement complexity of high dynamic range (HDR) imaging systems.

Paper accepted for Asilomar Conference on Signals, Systems, and Computers, November 2017.

Find out more in arXiv version (pdf)

Poster (pdf)

project name

Modeling Multidimensional Risk in Real-World Drivers with Diabetes

Goal is to detect driver glycemic state and predict associated driving risk using wearable and in-vehicle sensor measurements of driver physiology, health, and behavior. In order to achieve this, in-vehicle systems, wearable sensors, and procedures capable of quantifying real-world driving behavior in drivers with type 2 (T2) diabetes mellitus (DM) are deployed. Driver’s personal profiles of health and physiology before and during driving is determined that quantify the likelihood of health and safety relevant changes in real-world driver behavior and performance.

Multiple papers to be appeared at FAST Zero 2019, and AAAM, 2019.

project name

Answer sentence selection

In information retrieval, an open domain question answering system aims at returning an answer in response to the user’s question. We focus on answer sentence selection which involves identifying sentences that contain answer to a given question with a system that can learn the process using semantic knowledge available from the internet.

Find out more in detailed project report (pdf)

project name

Zero-shot learning for Image classification task

The work focuses on applying learned knowledge of multiple neural networks, trained to recognize different imageclasses, to improve on zero-shot inference performance (performance on classes unknown to the classifier) of the system as a whole.

Find out more in detailed project report (pdf)

project name

Head Motion Detection and Tracking for gaming control

The goal of the project was to control a javascript based Pacmen game by detecting and tracking the head movements. Implemented using OpenCV Library in Python, the algorithm uses Haar feature-based cascade classifiers to detect movement of head precisely in real-time. The project was exhibited at Electronics Section in annual techno-cultural exhibition, Shrishti, 2015.

Find out more in detailed project report (pdf)

Summer Internships

project name

Edge-based Registration for 3D morphing of facial expressions from image data

Guide: Patrick Horain, Telecom SudParis, Paris. (2015)

The project aims to generate real-time 3D face animations directly from images or videos. I helped in development of a system to morph 3D face model to a facial expression provided by RGB image. Blend shape model was used to describe the human face mathematically. Non-commercial library FaceTracker was used for initial estimation of 8 camera and 55 blend shape parameters.

Find out more in detailed project report (pdf)

project name

Optimizing the motion and the vision system of humanoids for robo-soccer games

Guide: Dr. Siti Norul Huda, National University of Malaysia, Kuala Lumpur. (2014)

Robot soccer game uses global camera to track the ball and humanoids. The obtained data is used to take decisions. Task was to improve accuracy and speed in terms of color detection, ball tracking and motion prediction. To improve ball tracking and prediction system of global camera and motion gaits of Humanoid, system was developed using OpenCV, Visual C++ & RoboPlus software package by Robotis. Morphological operations (Erosion and Dilation) were successfully implemented along with Image moments function using OpenCV for highly improved accuracy and detection. Motion gaits of humanoid robot were improved to realize Human-like walking based on CoG calculations using Roboplus software module.

project name

Textural Analysis for Land cover classification of Radarsat-2 Data

Guide: Dr. Dipanwita Haldar,SAC, Indian Space Research Organization (ISRO) (2013)

Different land-covers such as urban, river, rice crops, wet land etc. are to be identified and tagged from sattelite imagary. I processed 3 channel, 8 bit, dual polarized SAR data acquired by RADARSAT 2 by filtering it using speckle filters and enhancing it using Image enhancement algorithms.

Find out more in detailed project report (pdf)


Program and Outreach Coordinator -Data Science Reading Group (2016 - current)

Reading group at Data Science Lab that organizes weekly seminars on variety of topics related to Data Science.

Co-ordinator & Mentor -Institute Academic Reinforcement Program (2014 - 2016)

Program aims to help freshman students for their academics by arranging extra teaching sessions and mentoring by senior year students. At present, Aid Teaching Environment Sciences and Communication Skills courses to more than 80 freshman students every weekend as a part of Institute Academic Reinforcement Program.

Manager, Mentorship Program -IEEE IIT Roorkee Students Branch (2014 - 2016)

Mentorship program aims to encourage freshman year students for academic research by organizing several workshops, seminars and short-term research projects under the guidance of student-mentors.

Co-ordinator - IEEE IIT Roorkee Students Branch & IEEE Special Interest Group (2014 - 2016)

Working with team of 2 coordinators, 4 executives and more than 20 participants to Conduct weekly meets with activities like collectively reading review papers, sharing technical skills, presentations over different fields in EE with a vision of advancing technology for society. Regular Interaction with professors, graduate students to increase research awareness among undergrads.

Front-end Designer & Web-developer - Information Management Group, IIT Roorkee (2012 - 2013)

IMG is the group of 40 students of IIT Roorkee, for development and maintenance of the IIT Roorkee Intranet & Internet system. Group developed several intranet based web applications as well as Institute Website. These applications are entirely designed, programmed and maintained only by the IMG members.