Enhanced reflections in neural rendering of signed distance functions
Event details
| Date | 15.06.2023 |
| Hour | 09:00 › 11:00 |
| Speaker | Corentin Dumery |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Wenzel Jakob
Thesis advisor: Prof. Pascal Fua
Co-examiner: Prof. Alexandre Alahi
Abstract
While significant progress has been made in reconstructing 3D scenes, the ability to interpret and interact with the reconstructed environment remains a challenge. Neural implicit surfaces, in particular, have emerged as powerful tools for reconstruction but pose difficulties in handling and interpreting.
Our proposed approach focuses on addressing this challenge by incorporating semantic knowledge into neural radiance fields. By developing a framework that can infer and incorporate semantic information, we aim to enable systems such as autonomous robots to effectively make sense of the reconstructed scene and interact with it.
Background papers
Exam president: Prof. Wenzel Jakob
Thesis advisor: Prof. Pascal Fua
Co-examiner: Prof. Alexandre Alahi
Abstract
While significant progress has been made in reconstructing 3D scenes, the ability to interpret and interact with the reconstructed environment remains a challenge. Neural implicit surfaces, in particular, have emerged as powerful tools for reconstruction but pose difficulties in handling and interpreting.
Our proposed approach focuses on addressing this challenge by incorporating semantic knowledge into neural radiance fields. By developing a framework that can infer and incorporate semantic information, we aim to enable systems such as autonomous robots to effectively make sense of the reconstructed scene and interact with it.
Background papers
- NeuS: Learning Neural Implicit Surfaces by Volume Rendering for Multi-view Reconstruction, NeurIPS 2021, Peng Wang, Lingjie Liu, Yuan Liu, Christian Theobalt, Taku Komura, Wenping Wang
- Differentiable Signed Distance Function Rendering, SIGGRAPH 2022, Delio Vicini, Sébastien Speierer, Wenzel Jakob (president of the jury)
- NeSF: Neural Semantic Fields for Generalizable Semantic Segmentation of 3D Scenes, TMLR22, Suhani Vora, Noha Radwan, Klaus Greff, Henning Meyer, Kyle Genova, Mehdi S. M. Sajjadi, Etienne Pot, Andrea Tagliasacchi, Daniel Duckworth
Practical information
- General public
- Free