Added Colab notebooks for easy training and generation

[Tylersuard]

Added 2 Colab notebooks, one for training and one for generation. colab_generate.py and colab_train.py are modified versions of pretrained_example.py and train.py respectively, modified to work with these 2 colab notebooks.

[Tylersuard]

Any movement on this? Please review and approve. I would be proud to say I contributed to this revolutionary repo. It opened my mind to the possibilities of AI.

[ahmedshingaly]

how can I create stylegan model with (1, 18, 512)
my stylegan model is creating shape (1, 12, 512) and I cannot find the latent space developed by Puzer because of shape difference

in more details:
my model produce shape with (1, 12, 512) using (https://github.com/NVlabs/stylegan)
but when I use stylegan encoder (https://github.com/Puzer/stylegan-encoder) to find latent space it requires (1, 18, 512), do you have any idea how can I produce (1, 18, 512) model shapes instead of (1, 12, 512)?

[ricshaw]

Thank you for providing these colab notebooks, helped me get started training my own models. Do you know if it is possible to condition on a particular class? And how to generate images for a specific class?

[Tylersuard]

@ahmedshingaly No I do not, sorry!

@ricshaw I'm glad they helped you! The thing about StyleGAN is that it's unsupervised, meaning there are no classes. If you want to generate a certain class of images, you would need to train on images consisting entirely of that class. I could be wrong though: I seem to remember a video online of them selecting certain features, but I am not sure how to do that. Sorry!

[ricshaw]

One more question: do you know how to continue training from an existing checkpoint?

[Tylersuard]

@ricshaw I'm sorry, but I don't. As near as I can tell, train.py looks for previous "runs" folders and then loads .pkl and .tfrecord files from there.

If you pay for colab pro, you can train for 24 hours straight.

One idea might be:

1. Train for 24 hours and save your pkl file and tfrecords, anything else you would need to train
2. Re-open the notebook in a fresh session. Add the tfrecords files and train again for just one epoch or so.
3. That should have created a "runs" folder. You can add your .pkl files to that... I think

This is just based on what I read in the training/misc.py file