Torsten Sattler was part of the team that has been awarded the CVPR 2024 Best Student Paper Award for the paper „Zehao Yu, Anpei Chen, Binbin Huang, Torsten Sattler, Andreas Geiger. Mip-Splatting: Alias-free 3D Gaussian Splatting“ at the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2024, held on 17-21 June in Seattle, USA. The CVPR 2024 committee selected 10 best papers from more than 11,500 submissions. The Best Paper Awards are among the most prestigious recognition a computer vision scientist can receive.
CVPR is one of the premier conferences in the field of computer vision and, according to Google Scholar, one of the top-3 venues in all of science. CVPR is the leading forum for cutting-edge research in computer vision, artificial intelligence (AI), machine learning (ML), augmented, virtual, and mixed reality (AR/VR/MR), deep learning, and other related fields. CVPR 2024 had over 12,000 attendees.
„We are honored to recognize the winners of the CVPR 2024 Best Paper Awards,“ said David Crandall, professor of computer science at Indiana University, Bloomington, Ind., and one of the program chairs ofCVPR 2024. „The 10 papers selected this year – double the number awarded in 2023 – are a testament to the continued growth of CVPR and the field and all the advances that lie ahead.“ Torsten commented “I feel very lucky to be part of this work. Receiving awards at CVPR is a rare honor and I am happy for all members of our team, and especially Zehao Yu, the PhD student who is the main author of the paper, that our work was so well-revceived”.
The paper focuses on the topic of creating photo-realistic 3D models of scenes from images. It builds upon 3D Gaussian Splatting, a recently proposed approach to highly efficient and photo-realistic 3D reconstruction. The paper introduces principled approaches to improve 3D Gaussian Splatting. By taking the scale of individual scene parts into account, the proposed approach enables aliasing-free rendering of the scene from a much wider range of novel viewpoints compared to the original Gaussian Splatting formulation. As a result, the proposed Mip-Splatting approach significantly outperforms current state-of-the-art methods in out-of-distribution scenarios where testing is performed at sampling rates different from training. This results in better generalization to camera positions outside the distribution and different zoom factors.
Torsten Sattler is a RICAIP Tenure Track Position Holder at CIIRC CTU (since 2020) and currently also a member of the Applied Algebra and Geometry Group of the Robotics and Machine Perception department. Previously, he worked at the Chalmers University of Technology in Gothenburg and at ETH Zurich after receiving his PhD from RWTH Aachen University. His main research interest is 3D computer vision, with a focus on 3D reconstruction and visual localization.
We congratulate Torsten Sattler and his co-authors on this important award and wish them continued success in their research careers.
