hi! result is bad! about stylegan2-ada HOT 3 OPEN

@anthonyyuan can you give some more details about the data that you used? E.g, you took a sub-sample of the original FFHQ (if so, how many images and what criteria did you use for your sample choice), etc.
If you used a custom data set, how did you obtain such data? Is there a large imbalance in your data (e.g, ratio of male to female/age/race)?

I ask because it might be a good indicator of how well the augmentation works dependent on the available data. I suspect that this works better if there is not such a high imbalance of the observed features (in this case, gender, age, race), but rather just a small data size.

I will also try it myself and inform of the observed results.

from stylegan2-ada.

I test this demo by your model and default var .I have not do train my data

from stylegan2-ada.

I meet the same problem！I just test this demo by your model and default var. Just by running the command below.

 python generate.py --outdir=out --trunc=1 --seeds=5000-7000 --network=pretrained/ffhq.pkl

from stylegan2-ada.