o Curated and Annotated a comprehensive dataset of 6000 images from major Ship Datasets like VOID, SGF, VAIS, MARVEL, MODD, MarDCT. o Devised and optimized, scalable and robust Object Detection, Localization, and Tracking models for determining precise location of objects, accurate distance measurement, and effective determination of velocity and orientation of target ships through Stereo Camera Setup.o Compared Stereo Camera System with LIDAR for distance measurement, with the Stereo system achieving a higher level of accuracy for farther distances by 10%.o Deployed the Detection, Tracking, and Localization model on a miniaturized ship and conducted testing using a wave simulator basin.

o Spearheaded the team as a machine learning research lead and collaborated in a cross-functional team with the Electronics, GUI, and Mechanical Teams.o Engineered 8-bit Super Quantized and Pruned Classification, Object Detection, and Segmentation models for detecting cracks and steel defect for Spresense microcontroller.o Fine Tuned Diffusion based generative models for acquiring synthetic images and expanding the dataset. o Developed a Polygon Coordinates to Bitmask Image conversion script, streamlining segmentation masks and reducing the manual annotation workload by 47%.o Optimized camera capture script rate from 560 ms/image to 30 ms/image (18x Faster). o Implemented two-way communication via web sockets on a server, where ESP32 transmits Base64-encoded image strings, IMU coordinates, and model predictions as JSON.

Domain: Machine Learning And Deep Learning