Ashwani Rathee

I'm a recent master's graduate from University of California, Santa Cruz, where I specialized in computer vision and visual computing for smart glasses and autonomous vehicles. At UC Santa Cruz, I was advised by Prof. Alex Pang and Prof. James Davis.

Email  /  LinkedIn  /  CV  /  Scholar  /  Twitter  /  Github

Projects / Research

I'm interested in computer vision, deep learning, and image processing. Most of my work is about understanding the physical world via vision. Some projects are highlighted.

Built a ROS 2-based RC car platform for robotics and vision experiments, integrating cameras, IMU, motor control, visual-inertial odometry, real-time perception, and multimodal AI on edge hardware.

Master's Project under supervision of Prof. Alex Pang and Prof. James Davis: Developed a multi-agent refinement framework for 3D object detection, leveraging worker-superviser loop to improve detection accuracy and robustness over SOTA methods like BoxerNet and EFM3D.

Built a GPT language model from scratch, progressing from neural-network fundamentals to tokenization, self-attention, multi-head attention, Transformer blocks, KV caching, training, and text generation.

An interactive quadrotor simulation for exploring UAV dynamics, trajectory tracking, control, and 3D flight behaviors. Developed for ECE 263 under Prof. Elkaim at UCSC.

Implemented a multi-modal contrastive learning framework using MNIST dataset, exploring the alignment of image, point cloud, and label representations in a shared embedding space.

A training-free multi-agent framework that iteratively segments, critiques, and refines masks to follow long and detailed labeling guidelines.

We develop a visualization tool for multi-modal medical imaging data, enabling interactive exploration of 3D medical imaging data with support for segmentation overlays and volume rendering. This poster was presented at IEEE VIS'2025.

We explore the use of Parameter-Efficient Fine-Tuning (PEFT) techniques to improve the performance of Medical LLMs for thyroid nodule classification. This project was developed during CSE 244C: Deep Learning for Advanced Computer Vision taught by Prof. Yuyin Zhou.

Broadcasting algorithms like best effort, reliable and uniform reliable broadcast developed during CSE 232: Distributed Systems taught by Prof. Mohsen Lesani.

Developed an interactive 3D graphics playground for exploring rendering, animation, lighting, and real-time scene interaction during the course CSE 160: Introduction to Computer Graphics under Prof. James Davis at UCSC.

BT-Autonet is an autoencoder-based network for denoising brain MRI, improving image quality across Gaussian, Rician, and Rayleigh noise.

Developed Julia image I/O tooling during Google Summer of Code (GSoC) '22, adding EXIF metadata reading support through ExifViewer.jl and GIF image support through GIFImages.jl.

Miscellanea

Website Template from Jon Barron.