1. Multiple input multiple output channel channel in CNN convolutional layer
Concatenate multiple channels
1.1 Multiple input and output channels
- Each output channel can recognize a special mode
- The input channel convolution kernel recognizes and combines the patterns in the input
1 x 1 convolution layer, which does not recognize spatial patterns, which is equivalent to the fusion channel, which is equivalent to being a fully connected layer MLP. Equal to 3 channels, reshape to a one-dimensional vector.
2. Handwritten code implementation from scratch
3. Introduction to pytorch implementation
reference
https://www.bilibili.com/video/BV1MB4y1F7of?p=1