how to define a fully convolution net?

I'm trying to build a FCN to segment image, so how to define my input layer using imageInputLayer function ,dose it seems like imageInputLayer(auto, auto, 3)?

[QU HAIFENG]

Maybe I don't describe the problem clearly, I want to input arbitrary size image to my convolution net, tkz for any suggestion.

[Matt J]

Yes! How?? It's shockingly difficult to get clarification on how to do this.

Answer 1

[QU HAIFENG]

I found answer from document --Semantic Segmentation Examples: "A semantic segmentation network starts with an imageInputLayer, which defines the smallest image size the network can process. Most semantic segmentation networks are fully convolutional, which means they can process images that are larger than the specified input size. Here, an image size of [32 32 3] is used for the network to process 64x64 RGB images."

[Matt J]

That's good to know, but the imageInputLayer performs operations on the inputs that are not convolutional or shift invariant. Doesn't that defeat the goal of achieving an FCN? Or did you turn off all the normalizations?