search

omertov / encoder4editing

Official implementation of "Designing an Encoder for StyleGAN Image Manipulation" (SIGGRAPH 2021) https://arxiv.org/abs/2102.02766
MIT License
945 stars 154 forks source link

Output is a Solid Color, Regardless of input #52

Closed axumkevar closed 3 years ago

axumkevar commented 3 years ago

I'm sure someone's had this issue before and I'm hoping it's an easy answer. I've got a jupyter notebook version forked from this that I've resolved all the dependencies as far as I can tell, but e4e is outputting a single brown box no matter what I do, either from the notebook or inference.py, no errors are output, it just only outputs brown boxes. I've tried increasing the image size in utils/common.py, only producing noise. I have since re-pulled the code, but still receiving brown boxes as output.

example

omertov commented 3 years ago

Hi @axumkevar, When I run the notebook no such thing is happening, are you using the pretrained checkpoints provided in this repo? perhaps there is a version mismatch between the installed packages and the ones in our e4e_env.yaml?

Best, Omer

axumkevar commented 3 years ago

Strangely enough Styleclip itself produced the correct output from the latents, however I ran into a bigger issue with memory. Closing this issue as I believe the memory constraints are likely a culprit here.

axumkevar commented 3 years ago

After moving to a device with a larger memory in gpu, and using the same docker image, this problem went away. Definitely Memory Related. Seems to require a device with 8GB+ VRAM