I am trying to understand how Training and Convolution works in deep learning., based on my researches I understand this : First, as input, there are 2 things that are coming: raw mask, and input image that has been masked. Applying a partial convolutional layer to those two is producing 2 things: A reduced mask, and a feature image. But I really couldn't understand what is happening in partial convolution to produce a feature image. I wanna learn that. My second question is concatenation. I mean there is concat going on in UNET but.. why? what is its purpose?
I am trying to understand how Training and Convolution works in deep learning., based on my researches I understand this : First, as input, there are 2 things that are coming: raw mask, and input image that has been masked. Applying a partial convolutional layer to those two is producing 2 things: A reduced mask, and a feature image. But I really couldn't understand what is happening in partial convolution to produce a feature image. I wanna learn that. My second question is concatenation. I mean there is concat going on in UNET but.. why? what is its purpose?