The meaning of two types

[JL-Cheng]

In the file NE_dn.m , you mentioned 2 types : ave and gph. However , I didn't see these words in your paper . So I want to know what's the meaning of these two types ?

[justinkhuang]

@wangboyunze can correct me if I am wrong, but I think that these are two different methods for normalizing the weighted adjacency matrix of the graph for the random walk. In a symmetric network, the 'ave' option would create a degree-normalized (in unweighted networks)or col-sum-normalized (for weighted networks) graph laplacian. But due to the nature of matrix multiplication, the resulting matrix would not be symmetric. The 'gph' option creates a symmetric version of the normalized graph laplacian for random walks. However, I believe the TransitionFields.m script converts both of these matrices into a DSM for the random walk steps described in the paper.

Wikipedia actually has a pretty nice page on the graph laplacians used in random walks.

[JL-Cheng]

Ok, and would you like to share the codes that compute the FV and VLAD encodings with me?

[Spirit-0814]

In the file NE_dn.m , you mentioned 2 types : ave and gph. However , I didn't see these words in your paper . So I want to know what's the meaning of these two types ?
你好 不知道你有没有运行这个代码 我使用作者提供的demo数据运行这个代码 文章中说Wt是个双随机矩阵 但得到的结果并不是一个双随机矩阵

[JL-Cheng]

I remember the Wt I got is a doubly-stochastic matrix.

[Spirit-0814]

I remember the Wt I got is a doubly-stochastic matrix.

是直接运行这个matlab的代码的吗？ 我直接使用作者提供的matlab代码和demo中的butterfly的数据，但只是得到一个对称的矩阵，并不是双随机矩阵