Disclaimer: This article is a blogger original article, follow the CC 4.0 BY-SA copyright agreement, reproduced, please attach the original source link and this statement.
RGB image
RGB3 input image has a Channel, assuming 100 filter, there are 100 outputs feature map, rather than 300. Because the number of output remains the number of convolution kernel 100, but when the output is added to three RGB channels and operation
Grayscale (grayscale) image
Only an image input channel, assuming 100 filter, 100 have feature map, and is obtained directly.
select
Why RGB -> Grayscale
- No significant effect on the color image, and the color is susceptible to the influence of light, causing a large change RGB
- 3channel into 1channel, and without adding operation, reduced amount of calculation
Why RGB
- When the color of the image functions, such as the distinction between red or eliminates the need to convert RGB gradation when the blue hat
image example
Gray scale image
RGB image