Thread by @ilkedemir

Last day of the main conference at #CVPR2020, and here are my non-scientifically chosen top five papers @CVPR for the last day. Visit them in the next 10(?) hours! (1/6)

1. High-Dimensional Convolutional Networks for Geometric Pattern Recognition

Authors: Christopher Choy, Junha Lee, René Ranftl, Jaesik Park, Vladlen Koltun

Correspondences in 2D/3D forms geometric structures in higher dim, let's segment those with ND convnets!

#CVPR2020

2. PointGMM: A Neural GMM Network for Point Clouds

Authors: Amir Hertz, Rana Hanocka, Raja Giryes, Daniel Cohen-Or

Constructing a hierarchical GMM by attentional split, and using the encoding for shape interpolation and generation.

#CVPR2020

3. Butterfly Transform: An Efficient FFT Based Neural Architecture Design

Keivan Alizadeh vahid, Anish Prabhu, Ali Farhadi, Mohammad Rastegari

Split data to K, combine ith element in each, recursively fuse in next layers, 1Dconv++!
(not distracted by "butterfly", no)

#CVPR2020

4. Shape correspondence using anisotropic Chebyshev spectral CNNs

Authors: Qinsong Li, Shengjun Liu, Ling Hu, Xinru Liu

Now you can apply anisotropic convolution to your pretty kittens using LB eigenspace + oriented spectral kernels!

#CVPR2020

5. Smooth Shells: Multi-Scale Shape Registration With Functional Maps

Marvin Eisenberger, Zorah Lähner, Daniel Cremers

Shell operator = low freq geo (extrinsic). Embed shapes in extrinsic x spectral feature space, then matching = linear transformation and fmaps.

#CVPR2020

Bonus (as always): For an overview of geometry and learning in 3D and beyond, you can check our report at @ipam_ucla. From 3D shapes, to graphs, optimal transport, and more.

ipam.ucla.edu/wp-content/upl…

Non-CVPR.

@threadreaderapp unroll