English

Instant Neural Graphics Primitives with a Multiresolution Hash Encoding

Computer Vision and Pattern Recognition 2022-05-05 v2 Graphics Machine Learning

Abstract

Neural graphics primitives, parameterized by fully connected neural networks, can be costly to train and evaluate. We reduce this cost with a versatile new input encoding that permits the use of a smaller network without sacrificing quality, thus significantly reducing the number of floating point and memory access operations: a small neural network is augmented by a multiresolution hash table of trainable feature vectors whose values are optimized through stochastic gradient descent. The multiresolution structure allows the network to disambiguate hash collisions, making for a simple architecture that is trivial to parallelize on modern GPUs. We leverage this parallelism by implementing the whole system using fully-fused CUDA kernels with a focus on minimizing wasted bandwidth and compute operations. We achieve a combined speedup of several orders of magnitude, enabling training of high-quality neural graphics primitives in a matter of seconds, and rendering in tens of milliseconds at a resolution of 1920 ⁣× ⁣1080{1920\!\times\!1080}.

Keywords

Cite

@article{arxiv.2201.05989,
  title  = {Instant Neural Graphics Primitives with a Multiresolution Hash Encoding},
  author = {Thomas Müller and Alex Evans and Christoph Schied and Alexander Keller},
  journal= {arXiv preprint arXiv:2201.05989},
  year   = {2022}
}

Comments

To appear in ACM Transactions on Graphics (SIGGRAPH 2022). 15 pages, 13 figures, 3 tables