search

sian-wusidi / StegoGAN

Official code for StegoGAN: Leveraging Steganography for Non-bijective Image-to-Image Translation
19 stars 0 forks source link

Question on Training Models for Bidirectional Image Translation #3

Open OneCutePig opened 3 months ago

OneCutePig commented 3 months ago

Thank you for the fantastic work on StegoGAN!

I noticed that netA and netB are different models in your code. If I want to perform image translation between modality A and modality B in both directions (A→B and B→A), do I need to train the models separately for each direction?

Thanks for your help!

sian-wusidi commented 3 months ago

both directions are trained together, similar as cycleGAN. netA and netB uses different architectures because one side contains the unmatchable features not present in the other set in our setting

OneCutePig commented 3 months ago

I got it, thank you for your detailed response!