generated images quality?

[YoelShoshan]

Thanks for sharing!

In the images that you generate, are you seeing similar quality to the paper?
(From what I understand, on the paper they use a larger dataset to train on.)

[artcg]

Hey,
They are pretty good but not quite as good as the paper (I have been generating 64x64 samples instead of 128x128) - I'm away from my GPU right now but will post some samples this weekend!

[YoelShoshan]

Sounds great :)

[malreddysid]

Hi,
How many epochs did you train for to get good results?

[fwiffo]

We trained for 400k to 450k steps total, but the results were pretty good after 200k steps. The remainder added just a little bit of sharpness and things like finer hair detail. May want to check issue #3 in carpedm20/BEGAN-tensorflow for some mistakes made in that version to see if you have anything similar.

[malreddysid]

I ran the implementation as is with default parameters. Could you please post some images?

[artcg]

Thanks for the tip @fwiffo!

I'll check that out

[agingdata]

Thanks,author!
I am suffer from some issues when trained a model and want to generate some samples simply run;
when I execute "python main.py --start-epoch=N add-epochs=0 --train=False"this line,some error appear.
can you share execute comand line?
where N is the checkpoint ,I have a little confused what it is a num or produced model? i have produced the model eg.BEGAN_64_64_0005.tfmod.data-00000-of-00001.I try to set "N" as "1or5",and "-train=False"when execute is a error,I replaced it with "train=0 or train=1",only this way,the code when execute is occor errors.
can you give some guide about you generate the samples?
thanks again!