👨💻 About Me
Hi, I'm Mudit Jain, an engineer at heart with expertise in 3D/2D Machine Learning, SLAM, Computer Vision, GPU programming, and Embedded Systems.
Currently, my work is centered on using Multimodal AI, particularly fusing LiDAR and camera data for 2D/3D static/dynamic object detection/tracking. My interests include applied ML and CV research, real-time SLAM systems, and advancing computer vision capabilities for autonomous vehicles.
📰 News & Updates
2025: Serving as a reviewer for IEEE IV 2025.
2024: Joined Qualcomm as Senior Deep Learning Engineer in Multimodal AI.
2024: Mentored a project at Google Summer of Code on 3D Reconstruction.
2024: Serving as a reviewer for WACV 2025.
2021: Joined Qualcomm as Senior Machine Learning Engineer, XR Research.
2021: Graduated with Master's degree from University of California San Diego.
2019: Started Master's in Electrical and Computer Engineering at UCSD.
2019: Joined DroneLab at UCSD as Graduate Student Researcher.
2017: Promoted to Embedded System Software Engineer II at NVIDIA.
2016: Joined NVIDIA as Embedded System Software Engineer I.
2016: Graduated with Bachelor's degree from BITS Pilani.
2016: Selected for Google Summer of Code as Developer for RTEMS.
2015: Joined NVIDIA as an Intern.
2014: Joined Srujana Innovation Center as an Intern.
2012: Started Bachelor's in Electronics and Communication at BITS Pilani.
🎓 Education
-
Specialization in Machine Learning and Data Science
Courses: Linear Algebra, Probability and Statistics, Statistical Learning, Visual Learning, Computer Vision I & III, GPU Programming, Deep Learning and Applications
-
Minor in Data Science with focus on Embedded Systems and Computer Vision
🛠️ Skills
💻 Programming Languages
C++ [6+ years]
Python [6+ years]
🧠 Technical Knowledge Domains
Multimodal Large Scale Deep Learning [Ray, Kubernetes, PyTorch]
Classical Computer Vision [C++, OpenCV]
3D Computer Vision
Machine Learning [PyTorch, JAX]
SLAM [ORB-SLAM, VINS Mono]
Non-linear Optimization [Eigen, g2o, ceres, GTSAM]
Bundle Adjustment
Camera Calibration
Pose Graph Optimization
IMU Preintegration
Bayesian Inference
Embedded Systems
SIMD Programming [CUDA]
Model Optimization [TensorRT]
3D Reconstruction [NeRFs, Gaussian Splatting]
2D/3D Object Detection
🚀 Projects
Implementation of object detection using DINOv2 self-supervised vision transformers, enabling state-of-the-art zero-shot detection capabilities.
Optimization methods for large-scale bundle adjustment in 3D reconstruction challenges, focusing on efficiency and scalability.
Monocular depth estimation techniques for 3D scene understanding from 2D images using deep learning approaches.
Custom CUDA Implementation for Multi-Agent Reinforcement Learning
Accelerated Q-table updates and reward policies for multi-agent Q-learning using CUDA, achieving 100% training accuracy and 99.8% test accuracy in under 4 minutes on a 46×46 grid with 512 agents.
University of California San Diego • Jan 2021 - Mar 2021
Speeding up Mario RL with Custom Torch C++ Extensions
Developed custom CUDA kernels for linear, pooling, ReLU, and convolutional layers to accelerate the training and inference of a CNN for a Double Q-learning based RL agent playing Mario.
University of California San Diego • Jan 2021 - Mar 2021
AlertWILDFire Plume Detection
Deployed an ensemble neural network model across 610 cameras in California for early wildfire detection, using modified MaskRCNN with focal loss and an EfficientNet-based segmentation model with SCSE attention.
Drone Lab - UCSD • Jul 2020 - Jan 2021
Domain Adaptation for Semantic Segmentation
Trained OCNet on Cityscapes dataset and used CycleGAN-based domain adaptation to generate real-world-like data from gaming data, improving model performance through expanded training data.
University of California San Diego • Sep 2019 - Dec 2019
Image Denoising using Deep CNNs
Implemented and compared DnCNN, UDnCNN, and DUDnCNN architectures for image denoising, achieving up to 99.85% accuracy with U-Net with dilated convolutions.
University of California San Diego • Sep 2019 - Dec 2019
3D Reconstruction of the Anterior Segment of the Eye
Developed image processing pipeline and GUI interface for 3D eye model reconstruction by projecting patterns on the anterior segment and applying PCA and object tracking techniques.
Srujana Innovation Center and MIT Media Labs
🏆 Honors & Awards
100% Tuition Scholarship
DroneLab, University of California San Diego
Full tuition coverage for exceptional research contributions and academic merit.
Telangana Overseas Scholarship
Government of Telangana • University of California San Diego
Prestigious government scholarship awarded to exceptional students for pursuing graduate studies abroad.
MCN Scholarship
Birla Institute of Technology and Science, Pilani
Merit-based scholarship recognizing academic excellence and leadership potential.
📝 Blogs
Comprehensive overview of Bird's-Eye View (BEV) perception techniques for autonomous driving, covering image-based, LiDAR-based, and multi-modal approaches.
Guide to mastering C++ programming language from fundamentals to advanced concepts, with practical examples and best practices for efficient code development.
Detailed exploration of attention mechanisms in modern vision transformers, analyzing their application in both 2D and 3D computer vision tasks.
Analysis of parameter-efficient adaptation methods for vision models, including visual adapters, prompt tuning, and task-specific fine-tuning approaches.