GO-Surf: Neural Feature Grid Optimization for Fast, High-Fidelity RGB-D Surface Reconstruction

Wang, Jingwen; Tymoteusz, Bleja,

doi:10.48550/arxiv.2206.14735

Cited by 2 publications

(2 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We implemented the 2D, and 3D customized CUDA kernel of the triple backward grid sampler that supports cosine, linear, and smoothstep kernel (Müller et al 2022) and third-order gradients u xxc , u yyc with second-order gradients (Wang, Bleja, and Agapito 2022). As a result, the runtime and the memory requirement were significantly reduced.…”

Section: Methodsmentioning

confidence: 99%

PIXEL: Physics-Informed Cell Representations for Fast and Accurate PDE Solvers

Kang

Byeonghyeon

Hong

et al. 2023

AAAI

View full text Add to dashboard Cite

With the increases in computational power and advances in machine learning, data-driven learning-based methods have gained significant attention in solving PDEs. Physics-informed neural networks (PINNs) have recently emerged and succeeded in various forward and inverse PDE problems thanks to their excellent properties, such as flexibility, mesh-free solutions, and unsupervised training. However, their slower convergence speed and relatively inaccurate solutions often limit their broader applicability in many science and engineering domains. This paper proposes a new kind of data-driven PDEs solver, physics-informed cell representations (PIXEL), elegantly combining classical numerical methods and learning-based approaches. We adopt a grid structure from the numerical methods to improve accuracy and convergence speed and overcome the spectral bias presented in PINNs. Moreover, the proposed method enjoys the same benefits in PINNs, e.g., using the same optimization frameworks to solve both forward and inverse PDE problems and readily enforcing PDE constraints with modern automatic differentiation techniques. We provide experimental results on various challenging PDEs that the original PINNs have struggled with and show that PIXEL achieves fast convergence speed and high accuracy. Project page: https://namgyukang.github.io/PIXEL/

show abstract

Section: Methodsmentioning

confidence: 99%

PIXEL: Physics-Informed Cell Representations for Fast and Accurate PDE Solvers

Kang

Byeonghyeon

Hong

et al. 2023

AAAI

View full text Add to dashboard Cite

show abstract

“…Following [27,39,48], our approach utilizes a hybrid neural SDF representation F g with multi-resolution feature grids and hash encoding for the efficient learning and rendering of scene surfaces. Given an input query position x, F g converts coordinate input x into a concatenated feature vector from the multi-resolution hash encoding sampled with trilinear interpolation (the encoding used in Instant-NGP [27]).…”

Section: Neural Surface Representationsmentioning

confidence: 99%