NeRF知识点:不同Contraction的数学表达
目录
Foward-facing 场景:NDC(Normalized Device Coordinate)1
360°,object-centric 场景: inverse-sphere warping1
MeRF: Piecewise-projective Contraction1
自由轨迹: Persepective warping1
Foward-facing 场景:NDC(Normalized Device Coordinate)
主要针对forward-facing场景,好处在于:
“Once we convert to the NDC ray, this allows us to simply sample t‘ linearly from 0 to 1 in order to get a linear sampling in disparity from n to ∞ in the original space.”
数学表达形式为:
360°,object-centric 场景: inverse-sphere warping
mip-NeRF 360 [Barron et al. 2022]
缺点在于使得ray-AABB intersection难以计算,这意味着难以使得网络跳过那些空的地方。
MeRF: Piecewise-projective Contraction
为了解决mip-nerf中的ray-AABB难以计算的问题,对contraction公式进行了改进:
自由轨迹: Persepective warping
提出一种通用的warping函数,使得能够适应多种相机轨迹形式。
参考文献链接:2303.15951.pdf (arxiv.org)