Does texture network contain holistic content constraint? about faster-high-res-neural-inpainting HOT 4 CLOSED

The texture network is optimized such that it respects the content of the prediction from the content network, and also respects the texture outside the hole. So basically there are two constraints: one is the prediction from the content network which constraints the global content, the other is the texture from outside the hole which constraints the textures. Is it clear?

from faster-high-res-neural-inpainting.

Yes,from you answer,I can draw a conclusion that the optimization(Equation 1) finished in the texture network.And the content network only provide an object for the texture network.Parameters in content network are invariable when the texture network work.Is that right?

from faster-high-res-neural-inpainting.

Yes,from you answer,I can draw a conclusion that the optimization(Equation 1) finished in the texture network.And the content network only provide an object for the texture network.Parameters in content network are invariable when the texture network work.Is that right?

You are 100% correct. The content network is fixed and do not change after the first step. It only gives the initial prediction as a constraint to the texture optimization.

from faster-high-res-neural-inpainting.

OK，thank you!

from faster-high-res-neural-inpainting.