Nerf 常见的问题是:如果输入的训练图像不足(nerf往往要100多张各个角度训练图像) 会导致一些错误的几何错误 潜在的原因是 缺乏对 空的空间/不透明的表面(empty space and opaque surfaces)有约束。 所以作者提出了一个损失, 利用 Depth 来 supervise Nerf。 现有的Nerf pipeline 都会先用SfM来估计 相机参数。其实S...
笔者个人认为,引入depth是一种常用的优化方法(如在MVS算法中的depth map),不足以作为一个novel的点,但如果把解决的问题从如何提升NERF的效果如何仅用少量图片就能达到不错的NERF效果,那这个就是另一个novel的设定了,若有兴趣可以阅读本文中related work中提到的NERF from few views的topic。 Method 从上述流程图可...
Our work is based on the state-of-the-art framework VolSDF, which models 3D scenes by signed distance functions (SDFs), since this is more applicable for surface reconstruction compared to the standard volumetric representation in vanilla NeRFs. For evaluatio...
Depth-supervised NeRF: Fewer Views and Faster Training for Free CVPR, 2022 Kangle Deng1,Andrew Liu2,Jun-Yan Zhu1,Deva Ramanan1,3, 1CMU,2Google,3Argo AI We propose DS-NeRF (Depth-supervised Neural Radiance Fields), a model for learning neural radiance fields that takes advantage of depth ...
Finally, we fix camera poses and employ a NeRF, however, without a neural network,for dense triangulation and geometric verification. Poses,depth adjustments, and triangulated sparse depths are our outputs. Forthe first time, we show self-supervision within 5 frames already benefits SoTA supervised...
Finally, we fix camera poses and employ a NeRF, however, without a neural network,for dense triangulation and geometric verification. Poses,depth adjustments, and triangulated sparse depths are our outputs. Forthe first time, we show self-supervision within 5 frames already benefits SoTA supervised...
在上一篇介绍引入深度优化NERF的文章(DS-NERF) 中,以colmap的稀疏点云分布(假设为高斯分布)为目标,使ray termination distribution逼近真实的点云分布,学习的过程用color loss 加 KL散度的 depth loss加以约束。这个idea直观且简单,但也有明显的上限约束,即当点云的深度不够准确时,depth的优化可能起不到预想的作用...