#input-channels

Question 1: Select all that apply

An ImageNet-pretrained CNN starts with a convolution expecting 3 input channels (RGB), and you want to feed it single-channel grayscale X-rays. Which approaches make the first layer's output exactly equal to feeding an RGB image whose three channels all hold the gray value? Select all that apply.

Repeat the gray channel three times and keep the pretrained first layer unchanged1
Replace the first layer by a 1-channel convolution whose kernel is the sum of the pretrained kernel over its R, G and B slices2
Replace the first layer by a 1-channel convolution whose kernel is the mean of the R, G and B slices3
Put the gray image in the R channel, fill G and B with zeros, and keep the pretrained layer4

Timed test