how does stride work in denosing neural network

5 次查看（过去 30 天）
konoha
konoha 2022-5-13
回答： konoha 2022-5-18
I am studying the convolutional network in this article, https://www.mathworks.com/help/deeplearning/ug/denoise-speech-using-deep-learning-networks.html
The input image size is [129 8]. There are 16 cnn layers, some have [9 1] kernel and some have [5 1]. Let use [9 1] kernel. In one computation, cnn will compute a dot product of 9 vertival pixels of the input image if I understand them correctly. The stride is set to [1 100], this mean the kernel moves 1 pixel vertivally and 100 pixels horizontally. However, the input is 129x8.

请先登录，再回答此问题。

回答（1 个）

konoha
konoha 2022-5-18
the first cnn layer has kernel [9 8]. This has scanned through the input matrix and output is a vector. a [1 100] works as designed, which move vertically. But the choice of moving 100 horizontally is remained unclear.

类别

在 Help Center 和 File Exchange 中查找有关 Deep Learning Toolbox 的更多信息

标签

产品


版本

R2020b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by