Hi, I'm Swapnil!
I am currently a Masters in Computer Science student at New York University, Courant. I have worked at
Ernst and Young Global Delivery Services on various topics related to Document AI and my interests lie primarily in the
fields of Reinforcement Learning, Computer Vision and Natural Language Processing.
Bachelor of Technology in Computer Science (with Hons in Robotic Vision)
Machine Learning Engineer
Masters in Computer Science, Courant Institute of Mathematical Sciences
Interested in starting a conversation? Feel free to drop me a mail at 500swapnil@gmail.com or swapnil.gupta@nyu.edu. You can also contact me via the social media handles mentioned below!
Created this Python program using OpenCV for detecting traffic signs and 2 Deep CNN Image classifiers in Tensorflow, 1 for detecting a traffic sign and 1 for classifying which traffic sign it is. This is part of the autonomous driving car. Built this while working at Robotics Research Center, under the guidance of Dr.Madhav Krishna.
I've gone through various layers of image processing to narrow down number of images to sent to the classifier.
Source Code for this program can be found here
Close ProjectThe classic nintendo bomberman game on a unix terminal using python. Features include colorful display, multiple levels, music, enemies of varying strengths and speeds, automatic movement of enemies in real time, and an organized code with modularity, encapsulation and inheritance.
Source Code for this program can be found here
Close ProjectA 3D WebGL game created using JavaScript. It showcases implementations of colours, lighting, textures, shadows, interactive controls and collsions etc. Built this game as a part of the Graphics Course.
Source Code for this program can be found here
Close ProjectA 3D OpenGL game created in C++. It showcases implementations of bounding boxes, power-ups, interactive controls and collsions etc. Built this game as a part of the Graphics Course.
Source Code for this program can be found here
Close Project
A Tensorflow model, which implements the Stacked Hourglass Network architecture with several residual layers and can be trained to detect keypoints of any object. It’s a dynamic network with customizability in number of stacked layers, image size, etc. Also supports multi-GPU training.
Source Code for this program can be found here
Close ProjectAn AI Bot capable of playing 16x16 variant of Extreme Tic-Tac-Toe created using Mini-Max and AlphaBeta pruning. The search algorithm used is MTD(f) algorithm. Zobrist Hashing has been used to identify board states that have occurred before. Uses heuristics to find best move when computation time is too much.
Source Code for this program can be found here
Close Project
A pure Tensorflow+Keras TPU trainable implementation of SSD (Single Shot MultiBox Detector) using different backbones of Efficientnet which can be replaced with any ImageNet backbone. Tested on the PASCAL_VOC dataset. Implemented several low level data preprocessing and augmentations in pure tf functions for faster computation.
Source Code for this program can be found here
Close Project
A pyTorch program for extracting essential features from multiple style images and adding them to a single image. Produces an image with multiple styles added to the input image. Made it as the final project for the course Statistical Methods in AI taught by Prof. Ravi Kiran at IIIT - Hyderabad.
Source Code for this program can be found here
Close Project
A python program for training an artificial player in the classic Flappy Bird game. Uses Genetic Algorithm (NeuroEvolution) including crossover and mutations on a population of players to generate better models.
Source Code for this program can be found here
Close Project
