Parameters for replicating quantitative results

LynnHo / AttGAN-Tensorflow

AttGAN: Facial Attribute Editing by Only Changing What You Want (IEEE TIP 2019)

MIT License

605 stars 135 forks source link

Parameters for replicating quantitative results #27

Closed e-271 closed 5 years ago

e-271 commented 5 years ago

Hi, thanks for your great work! I am trying to replicate your quantitative results, but currently only get about ~46% accuracy. What parameters do you use to get these results?

Currently I have: 200 epochs im_size=64 inject=0 skip_connections=1 reconstruction_loss=10

LynnHo commented 5 years ago

@e-271 You can try skip_connections=0 for 64-model. However, some details of the code are different from the paper (for example, the 64-model in paper use layer normalization), therefore, the results will still have a difference. Or you can refer to #11 for a 128 model with similar accuracy.