Patrick Noras

Github  /  LinkedIn  /  CV  /  Email

Biography

I'm a Masters's student in Computer Science at RPTU Kaiserslautern specializing in 3D Computer Vision. Alongside my studies, I work as a graduate research assistant at the German Research Center for Artificial Intelligence (DFKI) in the Augmented Vision group under the supervision of Prof. Didier Stricker.

Prior to this, I completed my B.Sc. in Computer Science at RPTU Kaiserslautern. During my studies, I was an undergraduate software developer at SmartFactory KL and a visiting graduate researcher at UNC Chapel Hill , where I worked under the mentorship of Prof. Roni Sengupta, supported by a DAAD-PROMOS scholarship.

Research

My primary research interests lie in image-based rendering and its applications in 3D reconstruction and inverse rendering, with the goal to develop methods that robustly recover accurate geometry, appearance, and material properties

Currently, my research focuses on sparse-view 3D reconstruction and inverse rendering under challenging input conditions, spanning object-centric, indoor, and outdoor scenes.

Featured Projects

GAINS

GAINS: Gaussian-based Inverse Rendering from Sparse Multi-View Captures

Patrick Noras, Jun Myeong Choi, Didier Stricker, Pieter Peers, Roni Sengupta

ECCV 2026

Paper Project

GAINS is a Gaussian-based inverse rendering framework that leverages learning-based priors to improve geometry and material recovery from sparse multi-view captures.

Monocular360° GS

Monocular360° GS

Presentation Demo

This project introduces Monocular360° GS, a method for monocular panoramic image-based rendering that generates parallax through partial ground-truth views and uses inpainting to handle occlusions, achieving improved results on real-world scenes.

Bachelorthesis

Performance and Accuracy Assessment of Nvidia's Omniverse Isaac Sim for Generating Synthetic Data from Real-world Scenarios

Thesis Presentation Code

I created this project as part of my undergraduate thesis to evaluate NVIDIA's Omniverse Isaac Sim for generating realistic synthetic stereo-camera images and LiDAR point clouds by comparing simulated sensor data to a real-world scene.