Evin Pinar Ornek: From Unseen Objects to Holistic 3D Scene Understanding

The talk given by Evin Pinar Ornek in KUIS AI Talks on January 7, 2024

Title: From Unseen Objects to Holistic 3D Scene Understanding

Abstract: Deep learning and its applications in computer vision have revolutionized machine perception, leading to significant advancements in robotics, AR/VR, and autonomous vehicles. However, accurately interpreting complex scenes in a semantically rich manner and generating context-aware environments remain challenging, arguably due to the inherent difficulties in our three-dimensional world. This talk introduces Evin's research on 3D scene understanding via compositional and holistic reasoning. Starting from unseen object recognition in 3D (C3D ECCV 2022, SupeRGB-D RA-L 2023, FoundPose ECCV 2024), novel methods are introduced to capture the entirety of 3D scenes via scene graphs for high-level reasoning tasks (4D-OR MICCAI 2022) and generative modeling (CommonScenes NeurIPS 2023).

Short Bio: Evin Pınar Örnek is a final year Ph.D. student at TUM in the CAMP Computer Vision Lab under the supervision of Dr. Federico Tombari. Her research focuses on indoor 3D scene understanding for robotics and AR/VR applications, zero-shot learning in 3D, scene representations through semantic scene graphs, and the relevant tasks in spatial reasoning and generative modeling. She received a Google unrestricted gift for researching zero-shot learning in 3D. During her PhD, she has completed internships at Apple, Meta Reality Labs, and Amazon Lab126, working on XR headsets and household robotics research. She completed her bachelor's degree at Bogazici University Computer Engineering while spending time at Google, KU Leuven, and MPI for Intelligent Systems.