Dynamic Graph CNN for Learning on Point Clouds, Wang, Sun, Liu, Sarma, Bronstein, Solomon; 2018 - Summary
author: zayne-sprague
score: 8 / 10

The Core Idea of Dynamic Graph CNN

What are edges in a point cloud?

Edge Construction

How do we calculate the values of these edges (how is it realized technically)

\[x^′_i\ =\ □_{j:(i,j)∈\varepsilon}\ hΘ(x_i , x_j)\]

Edge Conv

Okay… but why is it called “Dynamic Graph”

How is this different from PointNet & PointNet++

Results of DGCNN

Specific Classification Archeticture

Classification Results

Segmentation Examples

What are the limitations?

Dependencies on Dense Data

The K hyperparameter table