Projects

1. Variable Quantized Ensemble Networks

   • Developed a system geared towards real-time semantic segmentation through variable quantization of ensembles of neural networks, trained on pre-determined groups of classes of Cityscapes Dataset and Indian Driving Dataset
   • Applied proposed method to state-of-the-art semantic segmentation models PSPNet and Deeplab-v3+, showing a 10% accuracy trade-off for large improvements in inference time and almost 20% reduction in memory usage
   • Developed using PyTorch and Tensorflow in Python

2. Distributed TicTacToe and chat room

   • Developed a distributed TicTacToe game, in a client-server setup using the Java RMI protocol
   • Parallelized the game server to handle multiple games, along with a chat server for multiple clients/client groups

3. AngelSafe

   • Designed and implemented a website focused on promoting safety of women during travel by making use of crowdsourced data to generate heat maps representing threat levels in a particular area
   • Secured first place in the Code.Fun.Do hackathon online round organized by Microsoft India in Hyderabad
   • Built using the MEAN Stack, consisting of MongoDB, Express.js, Node.js and Angular.js

4. Music Genre Classification

   • This project involved analysis of songs and predicting the genre using Neural Networks, SVM’s, KNN’s and Random Forests.
   • Performed analysis of songs and built feature sets using Mel Frequency Cepstral Coefficients (MFCC) of each track.
   • Implementation was done using Python, Keras and scikit-learn.
   • Neural Network approach gave better overall accuracy than the results published in the paper this project was based on.

5. Seeing the Arrow of Time

   • A project of the Computer Vision course to implement a paper titled Seeing the Arrow of Time which was submitted to CVPR in 2014, with the aim of being able to predict the direction of flow of time in a video.
   • Method was based on analyzing patterns of motion for temporal asymmetries, captured using a SIFT-like descriptor.
   • Implemented using Python and Numpy

6. Mini SQL Engine

   • Built an SQL engine in Python to parse and execute SQL commands, with relevant error handling

7. Model Compression Project in collaboration with Microsoft Research India

   • Involved compression of models(for video and image analysis related tasks), exporting them to mobile and testing performance, accuracy and hardware load
   • Built an Android app using Tensorflow Lite to run compressed models for detecting and processing actions and facial expressions of drivers while driving.

8. Privacy in Online Social Media

   • Course project with the aim of displaying the power that is possessed by social media companies, in the form of predictive analysis, with applications in targeted advertisements and a lot more.
   • Data of consenting participants was collected using publicly available API’s and Selenium
   • Analysis of data was done to generate statistics like daily active hours, frequently referenced topics, frequently used keywords, emotions in speech, closest friends and so on.

9. Game Development

   • Designed and developed a game in Unity3D, incorporating particle effects, projectile physics and collider mechanics
   • Developed variations of the popular games BrickBreaker and Bloxorz in 2D and 3D respectively using OpenGL
   • Developed an Aquarium model using WebGL

10. Seeing the Arrow of Time

    • A project of the Computer Vision course to implement a paper titled Seeing the Arrow of Time which was submitted to CVPR in 2014, with the aim of being able to predict the direction of flow of time in a video.
    • Method was based on analyzing patterns of motion for temporal asymmetries, captured using a SIFT-like descriptor.
    • Implemented using Python and Numpy