PET-NeuS: Positional Encoding Triplanes for Neural Surfaces

Yiqun Wang, Ivan Skorokhodov, Peter Wonka

Keywords: Multi-view Surface Reconstruction, Neural Radiance Fields, Signed Distance Functions

TL;DR: We improve NeuS by introducing Tri-planes, modulated positional encoding, and learned self-attention convolutions.

Abstract: The signed distance function (SDF) represented by an MLP network is commonly used for multi-view neural surface reconstruction. We build on the successful recent method NeuS to extend it by three new components. The first component is to borrow the Tri-plane representation from EG3D and represent signed distance fields as a mixture of tri-planes and MLPs instead of representing it with MLPs only. Discretizing the scene space with Tri-planes leads to a more expressive data structure but involving tri-planes will introduce noise due to discrete discontinuities. The second component is to use a new type of positional encoding with learnable weights to combat noise in the reconstruction process. We divide the features in the tri-plane into multiple frequency bands and modulate them with sin and cos functions of different frequency. The third component is to use learnable convolution operations on the tri-plane features using self-attention convolution to produce features with different frequency. The experiments show that PET-NeuS achieves high-fidelity surface reconstruction on standard datasets. Following previous work and using the Chamfer metric as the most important way to measure surface reconstruction quality, we are able to improve upon the NeuS baseline by 25\% on Nerf-synthetic (0.84 compared to 1.12) and by 14\% on DTU (0.75 compared to 0.87). The qualitative evaluation reveals how our method can better control the interference of high-frequency noise.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Deep Learning and representational learning