YUV pixel format for CV applications

Nowadays, for most of the vision related neural network processing, YUV pixel format is preferred over RGBA on video frames. Using of YUV pixel format instead of RGBA pixel format provides some advantages.

For sending 1920x1080 frame at 30FPS, just by using YUV420 pixel format we can get ~90 MBPS (36% of RGBA).

In YUV or YCbCr, Y represents the luminance component, which indicates the light intensity of the color. So we can get directly gray scale with luma channel alone.

YUV in vision related Neural network:

Current scenario for CV related neural networks:

So by using YUV pixel format in NN processing leads to increase in speed and decrease in memory.