How to predict label and θ when testing

[c1a1o1]

How to predict label and θ when testing

[elvisyjlin]

To predict labels from D, you can achieve it by

import torch
_, logits = attgan.D(img)
predictions = torch.sigmoid(logits)

As for the θ, Extension for Attribute Style Manipulation is not implemented. You will find no θ here.

However, in my implementation, only weights of G is saved. It is not feasible to predict labels from my pretrained weights. You have to change this line from saveG() to save() and train a model yourself to get a good D.

[c1a1o1]

Thank you very much!