Ajit Sarkaar

Graduate Student in Computer Vision and Deep Learning

Virginia Tech

About

Hello friend! Nice to see you here!

Let me tell you a bit about myself. I am a masters student in The Bradley Department of Electrical and Computer Engineering at Virginia Tech conducting research in Computer Vision using Deep Learning architectures under the guidance of Dr. A.L. Abbott. I work on solving problems related to panoptic image segmentation and scene understanding which have applications across a broad range of domains. I worked as a Research Intern at Amazon last summer and have TA’d for some highly popular courses such as Computer Vision and Artificial Intelligence at Virginia Tech. When I’m not in front of a computer, I play semi-professional soccer, click some photos and explore the exciting field of Cosmology. I hope to experience interstellar travel someday (hope :P). If you’re visiting this page, it is highly likely that we share some common interest and could help each other. I’d love it if we connected via some social channel. My info is to the left.

Look forward to meeting you! Cheers!

Interests

Computer Vision
Deep Learning
Artificial Intelligence
Signal and Image Processing

Education

Masters in Computer Engineering, 2020

Virginia Tech
Bachelors in Electrical and Computer Engineering, 2016

University of Pune

Skills

Programming Languages

Python: ⭐⭐⭐⭐⭐

C++: ⭐⭐⭐

MATLAB: ⭐⭐⭐⭐

Java: ⭐⭐

Swift: ⭐⭐⭐

JavaScript: ⭐⭐⭐

SQL: ⭐⭐⭐

Frameworks

NumPy: ⭐⭐⭐⭐⭐

OpenCV: ⭐⭐⭐⭐⭐

PyTorch: ⭐⭐⭐⭐⭐

TensorFlow: ⭐⭐

MXNet: ⭐⭐⭐

ROS: ⭐⭐

LaTeX: ⭐⭐

Tools

Git: ⭐⭐⭐⭐

Sublime: ⭐⭐⭐⭐⭐

PyCharm: ⭐⭐⭐

Eclipse: ⭐⭐⭐

XCode: ⭐⭐⭐⭐

Pentaho DI: ⭐⭐⭐

Unity: ⭐⭐

Research & Work Experience

Deep Learning Graduate Research Associate

CESCA, Virginia Tech

August 2018 – Present Blacksburg, VA

My research is focused on dealing with occlusion to improve performance on the panoptic segmentation task, organized jointly by COCO and LVIS. The learnings were also carried over to the perception stack of the autonomous car for the AutoDrive Challenge held by SAE and General Motors.

Research Scientist Intern

Amazon.com

June 2018 – August 2018 Seattle, WA

I was part of the autonomy team working on efficient computer vision algorithms for airborne object detection. Increased the computational efficiency by ~3X which enabled the object detection pipeline to run in real-time on a low resource computing platform.

Software Developer and Analyst

Merkle Sokrati (Dentsu Aegis Network)

March 2017 – July 2018 Pune, IN

Worked on the company’s tracking pixels product Chuknu and served as the primary stakeholder for the data engineering product Galaxy, serving over 30 clients.

Undergraduate Student Researcher

DSPIP Lab, MIT Pune

July 2015 – June 2016 Pune, IN

Designed a software synthesizer prototype for the musical instrument the Tabla, using Empirical Mode Decomposition and used neural network based classification and clustering algorithms for analysis of performance.

Projects

FaceNet: Learning an embedding for facial recognition, verification and clustering

This is an implementation of FaceNet (Google Research), a deep convolutional neural network based on the Zeiler and Fergus architecture to learn unified embeddings for faces in Python using Tensorflow and Keras.

Paraphrase Matching of Natural Language Questions

Quora released a dataset to detect duplicate questions on the platform. To solve sentence matching task, I came up with and implemented two deep recurrent architectures for semantic matching of sentences using LSTMs and Bi-LSTMs. The best model achieved a comparable F1-measure to SoTA for this task on the Quora question pairs dataset.

Pacman: Capture the Flag

Designed an artificial intelligence agent for the Pacman game which uses adversarial search techniques, reinforcement learning and bayesian inference to play the capture the flag against others. Was one of the top performers in the contest held for VTCS4804

Addle: Quiz your friends

Built and released a fun image processing application on the iOS app store written in Swift, which turns images into jigsaw puzzles and allows friends to play against each other. Allowed Facebook users to become friends and play against each other.

Teaching

ECE4554/5554: Computer Vision

I was a Teaching Assistant for the graduate course ECE4554/5554 Computer Vision with over 160 students. I help and evaluated students with core concepts and course assignments in Python and OpenCV during the Fall semester.

ECE4524: Artificial Intelligence

Also served as the TA for the course ECE4524 Artificial Intelligence and Engineering Applications for the Spring 2019 and Spring 2020 semesters. Responsible for the design of Python coding projects (based on UC Berkeley’s CS188 projects) and practical coursework for students