Hardcoded domain adaptation classifier in MUNIT/SPADE

[51N84D]

The domain adaptation classifier in the MUNIT/SPADE codebases (in "utils.py") doesn't work with arbitrary latent vectors.

For example, changing the number of downsampling layers breaks the code

[51N84D]

Temporary solution is modifying the architecture when changing the number of downsampling layers

[sashavor]

Which entails...?

…

On Fri., Apr. 10, 2020, 5:18 p.m. Sunand Raghupathi, < ***@***.***> wrote: The domain adaptation classifier in the MUNIT/SPADE codebases (in "utils.py") doesn't work with arbitrary latent vectors. For example, changing the number of downsampling layers breaks the code — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#125>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADMMIITNWGVRLIEBP2BXIWTRL6EKNANCNFSM4MFWW35A> .

[sashavor]

That is a lot of possible combinations though, right?

…

On Fri., Apr. 10, 2020, 5:19 p.m. Sunand Raghupathi, < ***@***.***> wrote: Temporary solution is modifying the architecture when changing the number of downsampling layers — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#125 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADMMIISPATROXM46VZRCUU3RL6EN7ANCNFSM4MFWW35A> .

[vict0rsch]

Some libraries forward 1 sample through networks to make this kind of decision. So one solution would be to give a sample to the networks creation procedure and infer the proper shapes given the sample and the config