Open travers-rhodes opened 3 years ago
Thanks for your pull request. It looks like this may be your first contribution to a Google open source project (if not, look below for help). Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).
:memo: Please visit https://cla.developers.google.com/ to sign.
Once you've signed (or fixed any issues), please reply here with @googlebot I signed it!
and we'll verify it.
ℹ️ Googlers: Go here for more info.
@googlebot I signed it!
Thank you so much for making this library!
I was trying to run baselines using this library's "conv_encoder" architecture for the VAE and noticed that that architecture (including in the publicly available models at, for example, https://storage.googleapis.com/disentanglement_lib/unsupervised_study_v1/0.zip) don't seem to match the architecture described in https://arxiv.org/pdf/1811.12359.pdf in Table 2 of the appendix.
In particular, it looks like the last two Conv2D embedding layers have a kernel of 2x2 instead of 4x4.
I have no idea if it makes any important difference, but my intuition is telling me that 4x4 might make more sense (as shown in the attached pull request), so will probably make that change on my end. I just wanted to reach out and see if you would possibly have time to check my thought process there.
Thanks so much!