kumarnilay27@gmail.com
I am a final year student at IIT Kharagpur pursuing Integrated Masters in Chemistry. I have worked on problems mostly related to Computer Vision & Deep Learning and their application in Robotics and Medicines.
This summer I did a research internship at Vector Institute for Artificial Intelligence in collaboration with Machine Learning Research Group at the University of Guelph, Canada as a Deep Learning Research Assistant where I worked on Synthesis of Single Domain Antibodies using Generative Networks. Before this, I interned at Lemnis Technologies in summer 2018 as a Computer Vision Intern where I worked on real-time Eye Tracking of user wearing a VR Headset, which was demonstrated at SIGGRAPH 2018.
Deep Learning Research Intern
Toronto, Canada • May - July 2019
The objective was to develop novel molecules using robust generative networks. I worked on modifying existing models and designing new models to generate new molecules, specifically Single Domain Antibodies and predict their properties by jointly training a Semi-Supervised Variational Autoencoders along with a deep MLP network. Also, I benchmarked the trained network on a supercluster by parallelly running multiple experiments on Salmonella Dataset.
Computer Vision Intern
Singapore • May - July 2018 • Demo Link
At Lemnis, I worked on Pupil Tracking and Glints Based Detection of Eyes using a gaze detection algorithm for a user wearing a VR headset. I incorporated varying focus of the human eye by working on a novel varifocal algorithm which enabled the eye-tracking in a VR headset. During my internship, I also worked towards resolving the FPS Barrier of OpenCV capture (i.e. restricted to 30 FPS) by using Direct Show method and even parallelizing capture and output in multiple threads, which improved the FPS by 400% on supported cameras. Moreover, the designed algorithms resulted in 800% reduction in the run time of the varifocal algorithm demonstration of the at SIGGRAPH 2018.
Deep Learning Intern
California, San Francisco • Remote • March - August 2018
During my internship at 223-D, I worked on various algorithms and applied them to real-world problems. I developed a blend of CEILNet for reflection removal from cars and Densenet for pixel-wise semantic segmentation of objects. Moreover, I performed object detection by using a mix of YOLO and Single Shot Multibox Detector and then extracted the object using Grab Cut algorithm. Thus, improving the image quality using VGG-19 on DPED dataset to generate DSLR like quality and hooked it to a server for real-time demo. Also, I produced a depth map from a single image by implementing FCRN-Depth Prediction algorithm and blurred the image according to depth.
My research interests broadly include machine learning and deep learning. In particular, I'm interested in research applicable to computer vision, robotics, medicines, and natural language understanding.
Inter IIT Tech Meet, 2019
IIT Bombay • Sept'18 - December'19 • Demo Link
As the Captain of the team and also a member of the vision team, I worked on computer vision algorithms for a robot capable of detecting fruits and navigating through farms to pluck the fruits autonomously. The bot can successfully navigate on a mapped path and detect fruits with the help of a computationally efficient binarized neural network along with a sliding-windows method, which is implementation on an FPGA board. The custom BNN-network has an accuracy of 95.8% on fruit identification and can run up to 30 FPS on an FPGA board. The bot uses point clouds and depth data acquired from X-box Kinect to extract the real-world coordinates of the detected fruits in 3D.
Winner, Inter IIT Tech Meet, 2018
IIT Madras • November'17 - January'18 • Demo of the Competition
My team won Gold Medal for the Automated Toilet Cleaning Robot Competition among 23 participating IITs all over India.
I worked on developing algorithms for a robot capable of navigating and identifying toilet seats and stains to clean the entire washroom autonomously. The bot uses Tiny-YOLO for commode detection on Raspberry-Pi using Movidius Neural Compute Stick, which is a Visual Processing Unit. For stain segmentation, I applied otsu binarization along with watershed algorithm, and for seat detection, I trained a HOG+SVM model.
IIT Kharagpur • March '16 - December'18 • Demo Link
The team secured World Rank 2 at 26th International Ground Vehicle Competition (IGVC) among the 43 participating teams worldwide. I worked on the model for detection of cars, identification of traffic lights and recognition of traffic signals. Implemented stereo-vision using PCL library, detected traffic light and traffic sign using faster-RCNN object detection algorithm.
KPIT Sparkle,2018 • Top 100 projects nationwide
Self-Sustainable Autonomous Water Cleaning Surface Vehicle (SeeSAW.CSV) got selected among top 100 projects from over 2300 projects submitted nationwide. SeeSAW.CSV is an autonomous robot that can detect and collect garbage in a lake, using vision and planning. I worked on the perception and path planning of the robot using Raspberry-Pi and developed a machine learning model for garbage detection.
IIT Kharagpur • December' 16
Successfully completed a 80+ hour IEEE certified workshop on Image Processing and implemented various Image Processing algorithms in OpenCV. Completed the final problem statement to solve the problem of keeping a track of number of goals scored in a Robo-soccer match and the number of people entering a room.
TCTD Challange, IIT Kharagpur
Led a team of 14 members from IIT Kharagpur for the annual Inter IIT Tech Meet, 2018-19.
Radhakrishnan Hall of Residence, IIT Kharagpur
Led a team of 40+ members from R.K. Hall of Residence in the Inter Hall Hardware Modelling 2018-19 for the prestigious annual Technology General Championship 2018-19.
Department Of Chemistry, IIT Kharagpur
Elected as the Department Representative for all the Undergraduate students of the Department of Chemistry. Member of the UG council for the session 2018-19, under the Dean of Undergraduate Studies, IIT Kharagpur
Website cloned from here.