Aug 2021 – Nov 2021 · 3 mos

Jun 2019 – Aug 2021 · 2 yrs 2 mos

• Optimising performance of Graph Neural Networks

Aug 2018 – Nov 2018 · 3 mos

• Generating Animated Version of Real Images Using Cycle GANs
• 1. Trained a GAN using three different datasets, Celeb-A dataset for real world images, Danbooru 2017 dataset foranimated images and an edge smoothened version of the Danbooru 2017 dataset.
• 2. Optimized a weighted loss function consisting of a coarse loss to retain the semantic content of the input imagesand a fine loss to preserve clear edges in the output animated images.•Obtained the coarse loss byVGG-16 Conv 53feature difference between real and transformed animated images.
• 3. Realized the fine loss using a discriminator by assigning the transformed variant and the edge smoothened variantas Fake and the animated images as Real.

May 2017 – May 2019 · 2 yrs

• M.Tech Thesis: Pedestrian Detection ’in the Wild’
• 1. Performed the first-ever pedestrian detector in the ’wild’ which can semantically differentiate walkers from twowheeler and three wheeler riders in a heterogeneous traffic environment prevalent in developing countries.
• 2. Employed Part affinity models to model and detect human joints for pose estimation; followed by a deep learnt classifier separating the pedestrians from the group of detected people.
• 3. Proposed a deep cascaded network using conditional decoding for counting of pedestrians in very dense crowds.
• 4. The network consists of a classifier for mapping the crowd into five classes based on its density and an encoder-decoder network which gives the density map conditioned on the output of the classifier
• 5. Published the work as 'MULTI-DIRECTIONAL PEDESTRIAN DETECTION AND FLOW MONITORING FROM TRAFFIC VIDEOS' at Transportation Research Board, Washington, 2019.

May 2018 – Jul 2018 · 2 mos · Bangalore

• 1. Developed a bandwidth optimized algorithm to correct the geometric distortion suffered by surround-view cameras in cars thus enabling enhanced Auto Parking feature.
• 2. Designed adaptive region based division of frames to account for the differential scaling in the distorted image.
• 3. Successfully improved bandwidth usage by70%resulting in its integration with the new System on Chip (SoC)

May 2017 – Jun 2017 · 1 mo

• Developed a standalone service in Java, for the extraction, validation and loading of files in database using property based configuration.
• Enhanced the service using Spring IOC and JDBC for integration with the live product EIG, built for a client of NRI Financial Solutions
• Facilitated the event driven file processing using Apache Camel Route and ActiveMQ and populated database using DAO and Entity bean.

Oct 2015 – Sep 2018 · 2 yrs 11 mos · IIT Kharagpur

• 1. Spearheaded a 50 member group for Data Science and AI related projects at IIT Kharagpur. Organized seminars and workshops in association with Two Roads Tech. Solutions, ZS Associates with400+ students participation.
• 2. Kharagpur Data Analytics Group is a student driven research group which aims at working on AI and Data Science projects. We also conduct workshops, seminars in the field of Data Science and Artificial Intelligence and also provide one-one mentorship to students.