I graduated with a B.Tech in Computer Science and Engineering from the International Institute of Information Technology, Hyderabad (IIIT-H) in 2014. As an undergraduate, I was an Honors student at the Center for Visual Information Technology (CVIT), advised by Dr. Anoop Namboodiri.
My research interests span the domains of Computer Vision, Vision based Human-Computer Interaction, Artificial Intelligence and Machine Learning. I am interested in giving machines the ability to visually understand their surroundings so that they can make accurate decisions.
In general, I am a curious person. I like to know and try out new things, be it technology, ideas, music, academics or food. The culture at IIIT-H has inculcated in me, a love for competitive programming. I usually go by the handle "yashdv" on most programming websites. Some of my handles are:
In my spare time, I like to play Table Tennis, Football and chess. Swimming is my favourite activity that I do at leisure.
Software Engineering Intern (Summer 2013)
Worked with the Geo-Imagery Team (Bangalore, India) and built a system to track Google's satellite imagery as it gets processed through its pipelines before being released publicly.
Web Development Intern (Summer 2012)
Designed and built an educational networking platform using the Drupal CMS. The main feature of this website is the courses portal and its services.
The India Digital Heritage Project is a joint effort by the
Govt. of India, academia and Microsoft research (India) to preserve the cultural
monuments of India and create 2D/3D user experiences of Indian heritage for the
general public. I built a 3D segmentation module to segment rich laser scanned
3D mesh models of monuments into its major components.
A system to detect and extract moves made in a game of chess as it is being played.
This was done as part of my Digital Image Processing project (in a team of 3) and
the challenge was to use purely image processing concepts in its implementation.
From the live video, we extract key frames representing the state of the board
between the moves. Using the concept of image subtraction, one can detect the
locations (to and from) where move has been made. For chess board calibration,
we use Hough Transform to locate boundaries and map locations in the image to
squares on the actual board. Finally, the chess moves are shown live on a
An image search engine that retrieved images similar to a queried image. Two
approaches were implemented. The first implementation uses the Bag of Words model.
In the second approach, we trained SVM and ANN classifiers on Global features
such as GIST and PHoG to classify images. The results were benchmarked on the
We stitch/merge images at their common regions to form a bigger picture. This is
done by estimating homographies between them using feature matching and a robust
estimator like RANSAC (RAndom SAmpling and Consensus). The result is analysed
using various detector-descriptor combinations like SIFT, SURF, ORB, MSER etc.