Adding examples for StyleGAN / StyleGAN2 about keras-io HOT 6 CLOSED

Hello @fchollet!

@Korusuke and I were working on the notebook for the StyleGAN2 implementation, and we had a question about how you want the notebook to be structured. Because many Keras examples focus a lot on the explanations, we thought it would be difficult to create a high-level description of the underlying networks behind StyleGAN. Additionally, the large amount of code required to train the network and the fact that neither Korusuke nor I can test the training code due to hardware restrictions has led us to a proposal.

Would it be okay if the notebook and python file just loaded a pretrained pickle file containing the weights for StyleGAN, and then just explained the generator part of it?

Another issue we ran into is the sheer amount of code required to make StyleGAN inferencing work. The code is roughly 700 lines, and because of the 300 line limit for examples, we weren't entirely sure how to proceed.

Please let me know when you get a moment!

from keras-io.

Would it be okay if the notebook and python file just loaded a pretrained pickle file containing the weights for StyleGAN, and then just explained the generator part of it?

The code should show the end-to-end workflow, including training. But when it comes to demonstrating the model (towards the end), you can just load an existing weights file. The purpose of the example is to enable someone else to get started with training their own version of the model on a new dataset (and then use it).

Another issue we ran into is the sheer amount of code required to make StyleGAN inferencing work. The code is roughly 700 lines, and because of the 300 line limit for examples, we weren't entirely sure how to proceed.

300 is not necessarily a hard limit if the code is already very concise. We could go up to 500. Over 500 is definitely not suitable for keras.io (and it should probably be a multi-file project).

Please try to make it as short and concise and readable as possible. If you still really cannot go below 500, then I would encourage you to publish as a standalone GitHub repo rather than a keras.io code example.

Thank you!

PS: prefer email to contact me, I don't get GitHub notifications

from keras-io.

Is there any update on this? Your name is still associated with the outstanding work item.

from keras-io.

Hi, Could you please update on this request.

Also, could you please take a look into the below StleGAN based examples published, Thanks!
https://keras.io/search.html?query=StyleGan

from keras-io.

This issue is stale because it has been open for 14 days with no activity. It will be closed if no further activity occurs. Thank you.

from keras-io.

This issue was closed because it has been inactive for 28 days. Please reopen if you'd like to work on this further.

from keras-io.