The objective of this project is to apply deep learning techniques to develop a detector that can find infrastructure objects in images. The Navlab goup has a large data base of road images taken from inside vehicles. Currently Navlab uses classic computer vision techniques to detect traffic signs and road damage.
We will develop a general detector based on state-of-the-art deep learning that should outperform the current detector and should be expandable to detect other infrastructure objects like fire hydrants or street lights.Fall 2016
I am currently working as a Teaching Assistant in the Machine Learning Department for the course "Topics in Deep Learning 10-807". Deep Learning is a new course being offered this semester by Prof. Ruslan Salakhutdinov.
My responsibilities as a TA are going to be grading assignments, framing questions for the assignments and exams, answering questions on piazza and during office hours.Fall 2016
NOAH is a company dedicated to providing data analytics solutions in Basketball to improve performance of players. I worked as an intern on doing exploratory computer vision work regarding the recognition of basketball players during basketball training and games. During this internship I have explored face verification using deep learning, player tracking, background subtraction and many other computer vision techniques.Summer 2016
The biometrics center in cyLab, CMU is set up to use biometric data in security applications. I am currently working as a Research Assistant. My role is to refactor code to C++ from python for a face recognition project which uses R-CNN's. Also, they are interested in taking my course project "Video Synopsis" forward and make it better.May 2016
I worked as an intern for a "Text to Speech project" jointly undertaken by PESIT and IIT Madras. My role was to perform a literature survery on current Machine Learning techniques used to predict prosody in speech. Prosody is the the patterns of stress and intonation in a language. It is the feature that seperates human speech from a monotonic robotic one. Prosody prediction is an unsolved problem in speech processing so I usually found myself reading research papers on current advancements in the time I spent there.Sep 2014 to Nov 2014
Li2 Innovations is a company which focuses on educating young engineers in the fields of robotics and embedded systems by providing workshops and courses. I took various courses with them entitled Speech Processing Robot, Embedded Systems with MSP430 and Arduino workshops. After the courses, I continued as an intern. My team was working on optimizing code for a 3-D printer. We also worked on using hand-gestures to make a robot change directions.Jun 2013 to Oct 2013