we really enjoyed reading your paper and your novel approach to anonymize face images by removing identifying information in the latent space of a face generation network and we appreciate you open-sourcing your code!
We would like to use it to anonymize our own images but find that the documentation seems to be incomplete.
Especially, since we start with an image instead of a latent vector as in your jupyter notebook, we would first need to use the encoder to transform our images to that space.
I have tried using the encoder object that is created as part of the StyleID class in styleid.py, however it seems to expect an input with 19 channels instead of the 3 of an RGB image. That let me to wonder whether that was actually the correct encoder to use.
I would really appreciate some guidance how to encode my own images to the latent space so that I can follow the guide in your notebook.
I think that ideally, your code would include a straightforward script that could be used to anonymize images directly, but that may be a future nice-to-have :)
Hi,
we really enjoyed reading your paper and your novel approach to anonymize face images by removing identifying information in the latent space of a face generation network and we appreciate you open-sourcing your code! We would like to use it to anonymize our own images but find that the documentation seems to be incomplete. Especially, since we start with an image instead of a latent vector as in your jupyter notebook, we would first need to use the encoder to transform our images to that space. I have tried using the encoder object that is created as part of the StyleID class in
styleid.py
, however it seems to expect an input with 19 channels instead of the 3 of an RGB image. That let me to wonder whether that was actually the correct encoder to use.I would really appreciate some guidance how to encode my own images to the latent space so that I can follow the guide in your notebook. I think that ideally, your code would include a straightforward script that could be used to anonymize images directly, but that may be a future nice-to-have :)
Thanks in advance Julian