davidbau / ganseeing Goto Github PK

I have some issues running the seeing_distribution notebook. It gives an import error
No module named '_prroi_pooling'

_prroi_pooling = load_extension('_prroi_pooling', [pjoin(root_dir, 'prroi_pooling_gpu.c'), pjoin(root_dir, 'prroi_pooling_gpu_impl.cu')],verbose=False)
at funtional.py.

I am not really sure if this is a version problem or something. Thanks!

Dear @davidbau ,

Could you remember which file or part of file in this repo related to the experiment 4.4 Layer-wise inversion vs other methods's (b) Direct learning of E in your paper ? It may learn a deep network E that inverts G directly, without layer-wise decomposition.

Thanks

Thanks for the great work!
But I am a little bit confused about the plots in figure 6. There are 3 rows of co-relation plots in it (shown as below). However, I don't see any explanation in the paper about how to interpret the x/y axis. Take the middle row for example, both the true and reconstructed layer4 feature are high-dimensional vectors. So what does the x/y axis means? Norm or something else?
Thanks in advance!

Dear David,

Thank you for sharing with us this wonderful work.

Could you tell me how can I get the inveted results from StyleGan? The repo seems to just contain ProgressiveGan.

Thank you for your help.

Best Wishes,

Alex