Key Projects
NeonTreeClassification Package
Google Summer of Code 2025
An open-source Python package for ecological research enabling multimodal tree species classification using NEON data. The package provides:
- Comprehensive dataset of 167 tree species with 47,971 individual tree crowns
- RGB imagery, 369-band hyperspectral data, and LiDAR canopy height models
- End-to-end data pipeline for automated downloading, cropping, and curation
- Pretrained classification models and preprocessing workflows
Multistream Architecture for Repetitive Action Counting
Created a multi-stream architecture utilizing RGB (using MoviNets backbone) and pose estimation (poses extracted using YOLOv7) techniques to accurately count the number of repetitions performed during exercise routines, as well as to effectively classify the specific exercises being executed.
Key Features:
- Multi-modal approach combining RGB video and pose estimation
- Real-time repetition counting
- Exercise classification capability
Indian Licence Plate Recognition System
Developed an Automatic Number Plate Recognition (ANPR) system for Indian license plates that classifies vehicle, plate type (back/front) and characters using a single model, achieving 25 fps performance. The dataset and code were made publicly available to facilitate future research.
Technical Approach:
- Two-stage model combining semantic segmentation for localization with LPRNet for character recognition
- Novel annotation technique to handle different plate orientations
- Released comprehensive dataset for Indian license plates in the wild
F1-tenth Self-Driving Car
September 2019 - December 2020
Implemented and designed self-driving algorithms including wall following, mapping & localization, and path planning using SLAM for a one-tenth model of an F1 car.
Technologies Used:
- LiDAR and stereo camera sensors
- ROS (Robot Operating System)
- NVIDIA Jetson TX-2 as SOC Board
- SLAM for mapping and localization
Based on the UPenn F1tenth project