Projects
Research and course projects in 3D computer vision, neural rendering and reinforcement learning.
I2DGS: Physically-Based Inverse Rendering with 2D Gaussian Representations
Technical University of Munich · Mar 2026
Two-stage inverse rendering with 2D Gaussian surfels and differentiable ray-traced shadows to recover relightable geometry and PBR materials from OLAT captures.
Framework for 3D Reconstruction Understanding
3D Computer Vision Practical Course, Technical University of Munich · Jan 2026
Benchmarking hybrid, feed-forward and self-supervised 3D reconstruction on ScanNet++, then probing how MapAnything uses its geometric inputs.
Seminar: GaussianDreamer — Fast Text-to-3D with Gaussian Splatting
Seminar on Neural Radiance Fields and Gaussian Splatting, Technical University of Munich · May 2025
Seminar report and talk reviewing GaussianDreamer (Yi et al.), which bridges 3D and 2D diffusion models to generate 3D Gaussians from text in about 15 minutes.
InstantSplat++: Registration-Free Gaussian Splatting from Sparse Image Sequences
Course project, Technical University of Munich · Feb 2025
SfM-free sparse-view Gaussian splatting that initializes from a single Spann3R point cloud and prunes redundant points with cumulative masking.
I-GaSp3R: Incremental Gaussian Splatting for 3D Reconstruction
ADL4CV course project, Technical University of Munich · Feb 2025
Pose-free, feed-forward 3D Gaussian splatting that reconstructs scenes incrementally from unposed image sequences at 21 FPS.
Comparison of Different State-of-the-Art Reinforcement Learning Methods in a 2D Simulation Environment
B.Sc. thesis, Real-Time Systems Group, Institute of Systems Engineering, Leibniz University Hannover · Apr 2023
Reinforcement learning for a UAV search-and-rescue task — seven agents, from PPO to DreamerV3, compared against a human operator.