Rohan Karnawat

Rohan Karnawat

Graduate Student Researcher

University of Southern California

I am a Computer Science grad student at USC Viterbi (Batch of 2021). Currently, I am working under Prof. Ram Nevatia as a Graduate Research Assistant affiliated to the IRIS Computer Vision LAB and the Information Sciences Institute. My project focuses on learning robust representations for defense against adversarial attacks in images.

I have previously worked as a Research Engineer in Samsung Research Institute, in the Advanced Technology Lab. I worked on three projects: Human-Object Interaction, AI Gallery and Iris-based Biometrics and Authentication.

I completed my Bachelors from IIIT Hyderabad in Computer Science and Engineering with an Honors from Computer Vision Institute of Technology under Dr Anoop Namboodiri. I have also been a Teaching Assistant for Computer Vision, Digital Signals Analysis and Statistical Methods in AI (Intro to Machine Learning).

I have explored several fields and have worked on a lot of projects individually as well as in teams. I have experience in full-stack web development, machine learning research, computer vision tasks and have a strong understanding of the foundations in Computer Science. I love taking part in hackathons and engaging myself in projects that challenge me.

I am actively looking for full-time Software Engineering or Research positions in the US, ready to start in the Summer of 2021.

Interests

  • Software Engineering
  • Full Stack Development
  • Machine Learning
  • Computer Vision
  • Artificial Intelligence
  • Data Science

Education

  • M.S. in Computer Science, 2021

    University of Southern California, Los Angeles

  • B.Tech. (Hons) in Computer Science & Engineering, 2018

    International Institute of Information Technology, Hyderabad

  • High School, 2014

    Indian School Muscat, Muscat

Experience

 
 
 
 
 

Graduate Research Assistant

USC IRIS Vision Lab

Mar 2020 – Present Los Angeles, California
Currently researching on methods to learn robust representations of images that can be defended against adversarially attacked / perturbed traffic sign and satellite images.
 
 
 
 
 

Research Engineer

Samsung

Jun 2018 – Aug 2019 Bangalore, India
  • Researched on and extended human object interaction in images to short videos. Incorporated deformable convolution for semantic role recognition.
  • Managed weekly releases for flagship mobile devices by handling post-processing, model updates, and error reporting.
  • Completed a commercial project for quick biometric based verification. Implemented a reflection based blur detection engine, performed quality testing and handled code analysis.
 
 
 
 
 

Software Developer Intern

Samsung

May 2017 – Aug 2017 Bangalore, India
Worked in the Intelligent Services department and was tasked with testing and comparative study of various deep learning libraries feasible for advanced speech recognition to port existing models on Kaldi.
 
 
 
 
 

Teaching Assistant

IIIT Hyderabad

Jan 2017 – May 2018 Hyderabad, India
Handled course content, assignments, and grading for the courses: Statistical Methods in AI, Computer Vision and Digital Signals
 
 
 
 
 

Web Developer

Digitant

Aug 2015 – Dec 2015 Hyderabad, India
Designed a page rank algorithm for subscribers to deploy recommendation engines on content publishing sites. Built a dashboard to facilitate tracking & traffic analysis.

Skills

Languages

C++
Python
C
HTML+CSS
MySQL
JavaScript

Tools

LaTeX
Bash
MATLAB
Android
Docker
Angular

Others

Tensorflow
Pytorch
OpenCV
CPlex

Major Projects

*
All Computer Vision Image Processing Deep Learning AI Python C++

Comic Strip Generation

An end-to-end model using Pix2Pix based C-GAN along with fine-tuned LSTM and experiments with VQA based joint embedding and InfoGAN to generate alternate endings to short Garfield comic strips
Code Dataset Project

C Shell

Developed a multithreaded command-prompt shell and basic kernel in C, with piping and regex enabled.
Code

Content Aware Image Resizing

Images retargeting using the Patch-matching approach based on an approximate nearest neighbor algorithm and bidirectional similarity
Code

Honors Research at Center for Visual Information Technology

Designed an expressiveness classifier from upper body pose (video) and audio recordings of meetings using a dual-input CNN. Generated indexed summaries by combining the 2 phases. Worked on developing a room description application for blind users.

Adversarial Attack on Satellite Images

Simulated a targeted and universal adversarial attack to misclassify satellite images of points of interest. Worked on building a robust defense against such attacks.
Code Dataset Project

Music Genre Detection and Mood Mapping

Used unsupervised clustering and a supervised classification ensemble to classify songs into 5 genres based on Mel Spectral features. Includes a mood based music recommendation engine
Code

Ultimate TicTacToe AI Bot

Developed an AI bot for Ultimate Tic-Tac-Toe and came up with a special heuristic for scoring states selected by an alpha-beta pruned minimax tree
Code

Video Summarizer

Created an interactive video synopsis player of a large directory various media that allows browsing and viewing
Code

Coursework

  • Courses at USC Viterbi: Algorithms, Deep Learning and its Applications, Multimedia Design, Database Systems, Advanced Computer Vision, Web Search and Information Retrieval

  • Technical Courses at IIIT-H: Digital Logic Processing, IT Workshop, Data Structures, Algorithms, Linear Algebra, Abstract Algebra, Group Theory, Basic Electronic Circuits, Operating Systems, Computer Systems Organization, Object Oriented Programming, Formal Methods, Engineering System Design, Digital Signals, Artificial Intelligence, Advanced Networks, Distributed Systems, Database Systems, Statistical Methods in AI, Image Processing, Computer Vision and Deep Learning, Information Security, Natural Language Processing, Optimization Methods, Music and Mind

  • Non-Technical Courses at IIIT-H: Nanobiotechnology, Biomolecular Architecture, Modern Science, Human Rights, Ethics, East-West Greco-Indian History, Intro to Humanities, Sculpture, Physical Education

*

Basic Room Description

Worked on a Blind Assistance application that creates a caption describing common objects in a room, on capturing its photograph.

Encrypted File Transfer (Networks and Distributed Systems)

Implemented TCP and UDP protocol, packet capture, weighted fair queuing, scheduling algorithms and IP Lookup over intranet. Used socket programming for serialised and encrypted data transfer.

Contrast Based Filtering for Salient Region Detection

Implemented the Saliency Filters paper [Perazzi et al., CVPR 2012] Performed two applicative tasks: Number Plate Extraction: Detected the number plate on cars based on contrasted coloring. Image Blending: Obtained a binary mask from the salient region for merging using pyramids.

Extractive Summarization

Extracted sentences of importance from CNN news articles to create a summary, using optimized unsupervised algorithms including Page rank and sequence learning with LSTM (Enc-Dec model). Compared the results with a preset abstractive summary using Pyrouge

Hair Stylist

Built a mobile application that assisted in deciding which hair style is best suited for the frontal face photo that would be captured on camera. Face features were extracted using Haar cascade and an SVM was trained to capture details.

Hostel Management Portal

Multifunctional web application using the MVC framework of Web2py.

Computer Graphics and OOP

Made a donkey kong simulator following concepts of Object Oriented Programming. Also developed a 3D maze runner game with a basic physics engine. Made a short movie with textures and voiceovers using OpenGL and Blender

Contact