Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
184 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
45 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

BANF: Band-limited Neural Fields for Levels of Detail Reconstruction (2404.13024v2)

Published 19 Apr 2024 in cs.CV and eess.IV

Abstract: Largely due to their implicit nature, neural fields lack a direct mechanism for filtering, as Fourier analysis from discrete signal processing is not directly applicable to these representations. Effective filtering of neural fields is critical to enable level-of-detail processing in downstream applications, and support operations that involve sampling the field on regular grids (e.g. marching cubes). Existing methods that attempt to decompose neural fields in the frequency domain either resort to heuristics or require extensive modifications to the neural field architecture. We show that via a simple modification, one can obtain neural fields that are low-pass filtered, and in turn show how this can be exploited to obtain a frequency decomposition of the entire signal. We demonstrate the validity of our technique by investigating level-of-detail reconstruction, and showing how coarser representations can be computed effectively.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (50)
  1. Vision-only robot navigation in a neural radiance world. CoRR, 2021.
  2. Pyramid methods in image processing. RCA engineer, 29(6):33–41, 1984.
  3. Ntire 2017 challenge on single image super-resolution: Dataset and study. CVPRW, 2017.
  4. Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields. ICCV, 2021.
  5. Stephen Boyd. Introduction to linear dynamical systems (course reader for ee263). https://ee263.stanford.edu/archive/263lectures_slides.pdf, 2007.
  6. The laplacian pyramid as a compact image code. TCOM, 1983.
  7. Tensorf: Tensorial radiance fields. ECCV, 2022.
  8. Eastman Kodak Company. Kodak lossless true color image. http://r0k.us/graphics/kodak/, 1999.
  9. Neural articulated shape approximation. ECCV, 2020.
  10. Multiplicative fourier level of detail. CVPR, 2023.
  11. Multiplicative filter networks. ICLR, 2021.
  12. Plenoxels: Radiance fields without neural networks. CVPR, 2022.
  13. nerf2nerf: Pairwise registration of neural radiance fields. ICRA, 2023.
  14. Instruct-nerf2nerf: Editing 3d scenes with instructions. CVPR, 2023.
  15. Sape: Spatially-adaptive progressive encoding for neural optimization. NeurIPS, 2021.
  16. Tri-miprf: Tri-mip representation for efficient anti-aliasing neural radiance fields. ICCV, 2023.
  17. Anil K Jain. Fundamentals of digital image processing. Prentice-Hall, Inc., 1989.
  18. Relu fields: The little non-linearity that could. SIGGRAPH.
  19. Adam: A method for stochastic optimization. ICLR, 2015.
  20. Panoptic Neural Fields: A Semantic Object-Aware Neural Scene Representation. CVPR, 2022.
  21. Mobilebrick: Building lego for 3d reconstruction on mobile devices. CVPR, 2023a.
  22. Neuralangelo: High-fidelity neural surface reconstruction. CVPR, 2023b.
  23. Bacon: Band-limited coordinate networks for multiscale scene representation. CVPR, 2022.
  24. Neural sparse voxel fields. NeurIPS, 2020.
  25. Marching cubes: A high resolution 3d surface construction algorithm. ACM, 1987.
  26. Level of Detail for 3D Graphics. Morgan Kaufmann Publishers Inc., 2002.
  27. Acorn: Adaptive coordinate networks for neural scene representation. SIGGRAPH, 2021.
  28. Occupancy networks: Learning 3d reconstruction in function space. CVPR, 2019.
  29. Nerf: Representing scenes as neural radiance fields for view synthesis. ECCV, 2020.
  30. Instant neural graphics primitives with a multiresolution hash encoding. SIGGRAPH, 2022.
  31. Sophocles J Orfanidis. Introduction to Signal Processing. Prentice-Hall, Inc., 1988.
  32. Deepsdf: Learning continuous signed distance functions for shape representation. CVPR, 2019.
  33. Dreamfusion: Text-to-3d using 2d diffusion. ICLR, 2023.
  34. On the spectral bias of neural networks. ICML, 2019.
  35. Miner: Multiscale implicit neural representations. ECCV, 2022.
  36. Residual multiplicative filter networks for multiscale reconstruction. NeurIPS, 2022.
  37. Implicit neural representations with periodic activation functions. NeurIPS, 2020.
  38. Direct voxel grid optimization: Super-fast convergence for radiance fields reconstruction. CVPR, 2022.
  39. Neural geometric level of detail: Real-time rendering with implicit 3D shapes. CVPR, 2021.
  40. Fourier features let networks learn high frequency functions in low dimensional domains. NeurIPS, 2020.
  41. Geoffrey Hinton Tijmen Tieleman. Lecture 6.5-rmsprop: Divide the gradient by a running average of its recent magnitude. COURSERA: Neural networks for machine learning, 2012.
  42. Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction. NeurIPS, 2021.
  43. Neural fourier filter bank. CVPR, 2023.
  44. Bungeenerf: Progressive neural radiance field for extreme multi-scale scene rendering. ECCV, 2022.
  45. Neural fields in visual computing and beyond. CGF, 2022.
  46. Polynomial neural fields for subband decomposition and manipulation. NeurIPS, 2022.
  47. Freenerf: Improving few-shot neural rendering with free frequency regularization. CVPR, 2023.
  48. Volume rendering of neural implicit surfaces. NeurIPS, 2021.
  49. Anti-aliased neural implicit surfaces with encoding level of detail. SIGGRAPH, 2023.
  50. Plenoctrees for real-time rendering of neural radiance fields. ICCV, 2021.
Citations (3)

Summary

We haven't generated a summary for this paper yet.

X Twitter Logo Streamline Icon: https://streamlinehq.com