Aetherion
2025
Aetherion is a Deep Q-Network agent trained to play laser hockey using reinforcement learning. This project implements a Dueling DQN architecture with Prioritized Experience Replay and self-play capabilities.
Aetherion
2025
Aetherion is a Deep Q-Network agent trained to play laser hockey using reinforcement learning. This project implements a Dueling DQN architecture with Prioritized Experience Replay and self-play capabilities.
SMPL Augmentation
2024
SMPLAug provides data augmentation for the widely-used SMPL human body model by empowering researchers to generate realistic, diverse SMPL body models at scale. Our innovative approach seamlessly integrates Kinect noise simulation, dynamic clothing generation, and symmetric pose variations, dramatically expanding both the quantity and quality of training data. This breakthrough enables robust 3D human understanding systems that perform reliably in real-world scenarios with limited data.
SMPL Augmentation
2024
SMPLAug provides data augmentation for the widely-used SMPL human body model by empowering researchers to generate realistic, diverse SMPL body models at scale. Our innovative approach seamlessly integrates Kinect noise simulation, dynamic clothing generation, and symmetric pose variations, dramatically expanding both the quantity and quality of training data. This breakthrough enables robust 3D human understanding systems that perform reliably in real-world scenarios with limited data.
SMPL2Biomechanics
2024
SMPL2Biomechanics is pipeline for estimating human biomechanical data (ground reaction force, joint torque) using the SMPL body model and AMASS motion datasets. The system employs inverse kinematics and inverse dynamics through the CusToM library to extract biomechanical parameters and biomechanical key points are mapped from the SKEL skeleton model. SMPL2Biomechanics enables comprehensive analysis and modeling of human motion dynamics at scale.
SMPL2Biomechanics
2024
SMPL2Biomechanics is pipeline for estimating human biomechanical data (ground reaction force, joint torque) using the SMPL body model and AMASS motion datasets. The system employs inverse kinematics and inverse dynamics through the CusToM library to extract biomechanical parameters and biomechanical key points are mapped from the SKEL skeleton model. SMPL2Biomechanics enables comprehensive analysis and modeling of human motion dynamics at scale.
Kinema-RGBD (Kinematic Mesh Articulation via RGBD)
2024
This project presents a few-shot framework for RGBD 3D registration of articulated objects. It constructs a kinematic template using $\sim10$ RGBD images and DINOv2 descriptors. Registration to a new instance is achieved via optimal transport matching and a subsequent optimization routine involving TEASER++. The proof-of-concept on humans yields sensible registrations but highlights the need for better kinematic constraints and geometry-aware descriptors.
Kinema-RGBD (Kinematic Mesh Articulation via RGBD)
2024
This project presents a few-shot framework for RGBD 3D registration of articulated objects. It constructs a kinematic template using $\sim10$ RGBD images and DINOv2 descriptors. Registration to a new instance is achieved via optimal transport matching and a subsequent optimization routine involving TEASER++. The proof-of-concept on humans yields sensible registrations but highlights the need for better kinematic constraints and geometry-aware descriptors.
Probabilistic 3D Reconstruction from Single RGBD Image
2024
This projects builds a probabilistic ensemble of 3D reconstructions of an object from a single RGBD image. It uses geometric primitives from a template mesh to generate a dataset, fits a Gaussian Process (GP), and samples from its posterior. This successfully captures and represents the geometric uncertainty in occluded regions.
Probabilistic 3D Reconstruction from Single RGBD Image
2024
This projects builds a probabilistic ensemble of 3D reconstructions of an object from a single RGBD image. It uses geometric primitives from a template mesh to generate a dataset, fits a Gaussian Process (GP), and samples from its posterior. This successfully captures and represents the geometric uncertainty in occluded regions.
Xsports Center
2022
Distributed microservices platform built with Spring Boot and Spring Cloud for managing sports facilities, lessons, user accounts, and reservations. Features service discovery via Eureka, unified API gateway routing, JWT-based authentication, and comprehensive quality gates including mutation testing, static analysis, and code coverage reporting. Designed for scalability, maintainability, and production deployment with independent service scaling, fault isolation, and CI/CD integration supporting automated testing and quality assurance workflows.
Xsports Center
2022
Distributed microservices platform built with Spring Boot and Spring Cloud for managing sports facilities, lessons, user accounts, and reservations. Features service discovery via Eureka, unified API gateway routing, JWT-based authentication, and comprehensive quality gates including mutation testing, static analysis, and code coverage reporting. Designed for scalability, maintainability, and production deployment with independent service scaling, fault isolation, and CI/CD integration supporting automated testing and quality assurance workflows.
UFO Ray Tracer
2021
UFO Ray Tracer is a modern ray tracing engine designed for photorealistic rendering. The system seamlessly integrates real-time rasterization preview with high-quality ray-traced rendering, enabling both interactive scene exploration and production-quality image generation. The implementation prioritizes performance through multi-threaded processing, advanced spatial acceleration structures, and highly optimized intersection algorithms.
UFO Ray Tracer
2021
UFO Ray Tracer is a modern ray tracing engine designed for photorealistic rendering. The system seamlessly integrates real-time rasterization preview with high-quality ray-traced rendering, enabling both interactive scene exploration and production-quality image generation. The implementation prioritizes performance through multi-threaded processing, advanced spatial acceleration structures, and highly optimized intersection algorithms.
QuBo
2021
QuBo is an interactive engagement platform designed to enhance communication in educational settings, particularly for online and hybrid learning environments. The platform organizes student questions through prioritized boards, enabling real-time interaction with live polls and pace monitoring. With secure session management, moderation tools, and anonymous participation options, QuBo facilitates structured dialogue between instructors and learners while maintaining engagement and responsiveness.
QuBo
2021
QuBo is an interactive engagement platform designed to enhance communication in educational settings, particularly for online and hybrid learning environments. The platform organizes student questions through prioritized boards, enabling real-time interaction with live polls and pace monitoring. With secure session management, moderation tools, and anonymous participation options, QuBo facilitates structured dialogue between instructors and learners while maintaining engagement and responsiveness.
Timemo
2020
Timemo Organiser is a Java-based desktop application designed to help users organize their tasks while leveraging the forgetting curve theory to optimize learning and memory retention. The application combines task management with a scientific approach to scheduling reviews and reminders.
Timemo
2020
Timemo Organiser is a Java-based desktop application designed to help users organize their tasks while leveraging the forgetting curve theory to optimize learning and memory retention. The application combines task management with a scientific approach to scheduling reviews and reminders.