CODEWITHSUNDEEP
conv-neural-networkrgbmonochrome
Paul Jurczak
Paul Jurczak

Reputation: 8183

What is the size difference between Convolutional Neural Networks that take as input an RGB vs. monochrome image?

VGG 16 Convolutional Neural Network takes an RGB image as an input, with 2^24 possible pixel values. It has 138,357,544 parameters.

enter image description here

Consider a minimum size version of this network for grayscale images, i.e. 256 possible pixel values, with all redundancies removed. Feeding (r, g, b) pixels with r=g=b to the original network should be used as an equivalency test. How many parameters this grayscale network would have? Would performance scale linearly with decreased number of parameters?

Upvotes: 1

Views: 42

Answers (0)

Related Questions