I want to use vgg16 for grayscale images with one channel

Hi Guyz.......i need to use CNN VGG16 for training my image samples but my images are gray scale and the input layer of VGG16 alexnet and VGG19 has 3 channels what can i do to resolve this issue. Please let me know some one.

Thanks,

[abhi abhi]

vgg16 and vgg19 using deep learning how can i design this architecture please help me

Answer 1

[Darshan Ramakant Bhat]

Since VGG16 is a pretrained model its input configuration cannot be changed.You can copy the first Chanel values to other two channel and create a 3 channel image out of your gray scale image.

rgbImage = cat(3, grayImage, grayImage, grayImage);

Give this image as the input to VGG16. I hope this will workout for you.

[Salma Hassan]

is this more computational time to duplicate the channel ...do you know any other method ?

[Image Analyst]

It simply allocates more memory for rgbImage. It should be extremely fast and the computation time should not really be a concern.

[Modar Alfadly]

You can do the same as suggested more efficiently by summing up the channels of the kernels of the first convolutional layer.