search

igul222 / improved_wgan_training

Code for reproducing experiments in "Improved Training of Wasserstein GANs"
MIT License
2.35k stars 669 forks source link

Inconsistency in interpolation. #40

Closed apisarek closed 7 years ago

apisarek commented 7 years ago

Hi! Awesome work :)

I've noticed that in all files (but no in gan_toy.py) you use interpolates = real_data + (alpha*differences) but not interpolates = (1-alpha)*real_data + (alpha*differences). I don't know if results in paper where trained with that inconsistency or if it changes something, but I just wanted to mention it and ask if this is a bug or not.

igul222 commented 7 years ago

They're exactly equivalent when alpha is drawn from U[0,1]

From: Andrzej Pisarek notifications@github.com Sent: Thursday, August 3, 2017 7:28:08 PM To: igul222/improved_wgan_training Cc: Subscribed Subject: [igul222/improved_wgan_training] Inconsistency in interpolation. (#40)

Hi! Awesome work :)

I've noticed that in all files (but no in gan_toy.py) you use interpolates = real_data + (alphadifferences) where but not interpolates = (1-alpha)real_data + (alpha*differences). I don't know if results in paper where trained with that inconsistency or if it changes something.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/igul222/improved_wgan_training/issues/40, or mute the threadhttps://github.com/notifications/unsubscribe-auth/ABBP7nIjT8UQZg0P7tuP65lbvVmDboL1ks5sUqvogaJpZM4OtSj3.

apisarek commented 7 years ago

Oh, sure, I've mistaken differences with fake_data somehow. Thank you for quick response.