Fast Learning Radiance Fields by Shooting Much Fewer Rays

Abstract

Learning radiance fields has shown remarkable results for novel view synthesis. The learning procedure usually costs lots of time, which motivates the latest methods to speed up the learning procedure by learning without neural networks or using more efficient data structures. However, these specially designed approaches do not work for most of radiance fields based methods. To resolve this issue, we introduce a general strategy to speed up the learning procedure for almost all radiance fields based methods. Our key idea is to reduce the redundancy by shooting much fewer rays in the multi-view volume rendering procedure which is the base for almost all radiance fields based methods. We find that shooting rays at pixels with dramatic color change not only significantly reduces the training burden but also barely affects the accuracy of the learned radiance fields. In addition, we also adaptively subdivide each view into a quadtree according to the average rendering error in each node in the tree, which makes us dynamically shoot more rays in more complex regions with larger rendering error. We evaluate our method with different radiance fields based methods under the widely used benchmarks. Experimental results show that our method achieves comparable accuracy to the state-of-the-art with much faster training.

Fighting the Omicron variant: experience in Shenzhen

“…For the first time, Mildenhall et al [39] propose the neural radiance field to implicitly represent static 3D scenes and synthesize novel views from multiple posed images. Inspired by their successes, a lot of NeRF-based models [2], [10], [12], [14], [20], [21], [22], [24], [26], [34], [36], [37], [40], [42], [44], [46], [49], [53], [55], [64], [67], [75], [78] have been proposed. For example, point-NeRF [65] and DS-NeRF [15] incorporate sparse 3D point cloud and depth information for eliminating the geometry ambiguity of NeRFs, achieving more accurate and efficient 3D point sampling as well as better rendering quality.…”

Learning Consistency-Aware Unsigned Distance Functions Progressively from Raw Point Clouds

“…Classic optimizationbased methods [15,3,28,29] tried to resolve this problem by inferring continuous surfaces from the geometry of point clouds. With the rapid development of deep learning [63,33,56,61,27,24,54,55,58,53], the neural networks have shown great potential in reconstructing 3D surfaces [30,9,8,22,13,17,51,43,31,52]. In the following, we will briefly review the studies of deep learning based methods.…”