目录

目录

PatchNets: Patch-Based Generalizable Deep Implicit 3D Shape Representations


<PatchNets> PatchNets: Patch-based generalizable deep implicit 3D shape representations

Motivation

  • mid-level patch-based SDF
  • 因为在patch层次,不同类别的物体有相似性,用上这种相似性就可以做更泛化的模型
  • 在一个canonical space下学到这些patch-based representation
  • 从ShapeNet的一个类别学出来的representation,可以用于表征任何一个其他类别的非常细节的shapes;并且可以用更少的shape来训练
  • https://longtimenohack.com/posts/paper_reading/2020eccv_tretschk_patchnets/image-20201217115545052.png

Overview

  • auto-decoder
  • losses:重建loss和patch extrinsics的guidance loss,还有regularization

extrinsic loss

  • 这个loss保证所有的patch都对surface有贡献,并且处于caonical space
  • i个物体的patch extrinsics: $\boldsymbol{e}i=[\boldsymbol{e}{i,0},\boldsymbol{e}{i,1},\ldots,\boldsymbol{e}{i,N_P-1}]$
  • $ \mathcal{L}_{ext}(\boldsymbol{e}i) = \mathcal{L}{sur}(\boldsymbol{e}i) + \mathcal{L}{cov}(\boldsymbol{e}i) + \mathcal{L}{rot}(\boldsymbol{e}i) + \mathcal{L}{scl}(\boldsymbol{e}i) + \mathcal{L}{var}(\boldsymbol{e}_i) $
  • $\mathcal{L}_{sur}(\boldsymbol{e}_i)$ 保证每个patch都离surface很近
    • $\underset{逐patch}{\max}[surface上的所有点到该patch距离的最小值]$
  • $\mathcal{L}_{cov}(\boldsymbol{e}_i)$ symmetric coverage loss,鼓励surface上的每个点都至少被一个patch涵盖
  • $\mathcal{L}_{rot}(\boldsymbol{e}_i)$ 把patches和surface normals对齐
  • $\mathcal{L}_{scl}(\boldsymbol{e}_i)$ 鼓励patches to be reasonably small,防止不同patch之间显著的重叠
  • $\mathcal{L}_{var}(\boldsymbol{e}_i)$ 鼓励所有patch大小相似

result

  • https://longtimenohack.com/posts/paper_reading/2020eccv_tretschk_patchnets/image-20201217122000974.png
  • https://longtimenohack.com/posts/paper_reading/2020eccv_tretschk_patchnets/image-20201217122618035.png