justinpinkney / awesome-pretrained-stylegan2 Goto Github PK

https://drive.google.com/uc?export=download&id=1-F-RMkbHUv_S_k-_olh43mu5rDUMGYKe

via: https://github.com/mfrashad/ClothingGAN

!gdown --id 1AHKufJ5o4dQFThFN0A-pJRRC2lzxEH2J

Source: https://towardsdatascience.com/ganscapes-using-ai-to-create-new-impressionist-paintings-d6af1cf94c56

From Gabriel Syme on the TPU podcast discord:

Hello everyone! Given that I'm still struggling to get colab TPUs going, and I don't think I'll be spending more credits on vastai for this right now, I decided to share my Stylegan2-ada-pytorch model trained on about 3,000 real and stylistic images of architectural buildings, across 18 diverse architectural practices. Class (practice) information was not used during training, this is a 512x512 model and has been trained for a bit over 3kimg. I will eventually work on building a larger data set in the future across more practices but for now this is it.

The model isn't as good as I would like it to be, but it can still generate quite nice images. Maybe someone finds a cool seed!

pkl file: https://storage.googleapis.com/pretrained-stylegan2/stylegan-ada-pytorch-buildings-3kimg.pkl
dataset: https://storage.googleapis.com/pretrained-stylegan2/Buildings.rar

This is a model from the new stylegan-ada-pytorch repo

Hi,

Firstly thanks for the repo, it is indeed awesome :)

A couple of questions:

• do you have a contact for @Krrrl, author of the lifedrawings dataset? I'm struggling to reach them.

• do you have any thoughts on ethical issues around using the lifedrawings dataset in 'professional' work? I'm a musician working on an audiovisual commission (aimed much more at a musical audience than a visual art audience) and I REALLY want to use that dataset for one sub-piece in it. I would be significantly altering the appearance of the images, but still using them without the author's permission feels a bit wrong. I was wondering if you knew of their intention in making the dataset publicly available, if there was any intention behind it all.

Best wishes,

Mark

Add details of if the network was trained using transfer learning, and if so what network it was from.

This will be useful for using the models for fun like interpolation and blending (e.g. https://www.justinpinkney.com/stylegan-network-blending/)

This will also be important when adding models in #7

https://github.com/jeffheaton?tab=repositories&q=pretrained&type=&language=
Minecraft 1024px, Fish 256px, 1970s Sci-Fi 1024px, Christmas 256px

Judging by the fakes000000.png images from each, they weren't using transfer learning.

https://github.com/joshchen984/pretrained-stylegan2-hedgehogs
Hedgehogs 256px

https://github.com/dobrosketchkun/NeuralKuvshinov
https://drive.google.com/drive/folders/1T6BNlyPpvLsxXI-gkCepR7S_nFcSi63o (larger number = trained longer)
kuvshinov_ilya instagram artwork, transfer trained from stylegan2-ffhq-config-f.pkl

this is quite good - I wonder how hard it would be to fire up this with swapped models.

https://github.com/stylegan-human/StyleGAN-Human

Hi, nice work. I wonder is there any pre-trained model for lsun bedroom and landscpae? Thanks a lot.

I am running a notebook called "structured dreaming". The notebook loads the pkl file using this:

print('Loading networks from "%s"...' % network_pkl) device = torch.device('cuda') with dnnlib.util.open_url(network_pkl) as f: G = legacy.load_network_pkl(f)['G_ema'].to(device) # type: ignore

For network_pkl, this works:

https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada-pytorch/pretrained/ffhq.pkl'

but when I try one of your pkl files, for instance:

https://github.com/justinpinkney/awesome-pretrained-stylegan2#faces-FFHQ-config-f-512x512 (after downloading it first and then putting it into Gdrive), I get:

`Loading networks from "/content/drive/MyDrive/pretrained/ffhq-512-avg-tpurun1.pkl"...

ValueError Traceback (most recent call last)

in ()
5 device = torch.device('cuda')
6 with dnnlib.util.open_url(network_pkl) as f:
----> 7 G = legacy.load_network_pkl(f)['G_ema'].to(device) # type: ignore
8 #
9 # Read the data from the file

1 frames

/content/stylegan2-ada-pytorch/legacy.py in convert_tf_generator(tf_G)
153 unknown_kwargs = list(set(tf_kwargs.keys()) - known_kwargs)
154 if len(unknown_kwargs) > 0:
--> 155 raise ValueError('Unknown TensorFlow kwarg', unknown_kwargs[0])
156
157 # Collect params.

ValueError: ('Unknown TensorFlow kwarg', 'resolution_h')

`

Do you have any suggestions? Your list of pretrained networks is great!

Can anyone tell me how to run the script to generate a new image? I've already downloaded the weights. How to create new images using GAN with the script?

https://thisbeachdoesnotexist.com/

Hi,
i want to resume training with car config-f, when I started the training after the first tick, if i preview the init fakes they have messed up colours. Is this normal or is there some conflict with a colour space I can choose somehow?
Thanks

Doron Adler shared a link to a bunch of fine tuned models i.e. transfer learning from a base model for just a few ticks.

https://twitter.com/Norod78/status/1255237713118343168

These often give weird but interesting results, doesn't quite feel right to put them in the main list, but there should be a link somewhere.

Hi,
I was wondering if you're aware of a pretrained 256x256 FFHQ StyleGANv2 model available? I think it would be really be helpful for tasks where the full model is too big to start with.

the generated images using the provided pretrained model in these links contain colorful specks. For example, dataset: faces (FFHQ config-f). Anybody knows what's wrong with this?

new edition from @stuartwaller

https://github.com/StuartWaller/gen-art-project

model
https://drive.google.com/file/d/1AUVmc1wYSJ01JFvqGa4Bb0OERLdA0fXg/view?usp=sharing

https://github.com/seawee1/Did-Somebody-Say-Corgi

Dear Justin,

Thank you for creating the great repo. Do you have plan to support stylegan3?

Thank you for your help.

Best Wishes,

Alex

Hi ,Really thanks for your work.
From your blog I notice that blending needs a base model and a fine-tuned model from that. So I try it and, I want to know that if there is a fine-tuned model for Anime Face in stylegan2 ?
The Aaron Gokaslan's anime mode is not fine-tuned and can't be used for blending with ffhq. So if it's possible to use a 512-f-ffhq model fine-tuning an anime-face dataset so that the new anime face model can be used for blending?

Generate some interpolation videos for the different models