0

In autoencoder structure, when encoder side (deformable conv. layers) and decoder side (transposed deformable(?) conv. layers) are seperated, i want to inform the decoder about the offsets used in the encoder side. Desiring result is for the decoder to perform inverse(like) transform of the encoder's deformable convolution layers.

Has anyone ever thought about this?

Thank you in advance.

김영웅
  • 1
  • If your question is along the lines of "how good of an idea is this?", "Has this been done before?", or "How to use this?" then it is clearly off-topic for Stack Overflow. However, if your question is about how to implement such operation and architecture then you need to provide detailed information on how it is to be implemented and show the effort you've put in to illustrate what you've tried: this should include a minimal example with input and desired outputs as well as some code at best. – Ivan Jun 29 '22 at 07:26
  • Thanks for your kindness. I will rewrite question with considering your advice! – 김영웅 Jun 30 '22 at 08:52

0 Answers0