Great work!

The unet prediction result will be cropped to be the same size as the training input, e.g. 256 or 512.
For arbitrary image size (e.g. 1280*720), how to config or set the model to output the same size of the input image as your colab did?
Thank you.

The operation _weight_norm is unsupported during convert to CoreML model, I've been stuck on this issue for over a month.

wonderful work! really a nice piece of code. My only worry is that the code rotates my videos. Is this expected? how can I fix that?

Thanks!

Добрый вечер.Извините,это опять я.Снова эта ошибка появляется.Можно ли,самому эту ошибку решать?Или исправлять можете только вы?Обьясните пожалуйста подробно.

Hello, I try your model and it's amazing, but I find in some pictures if the nose is too big, there will be eyes in the nose. I try to lower the 'target_face' and it can work. But the details like the light of the eyes and background will also lose when I lower the 'target_face'. So I wonder is there a way to prevent the eyes occurs in the nose and keep the details in the meantime?

Hi,
i'm really interested in repo and i want to run in my computer instead of hugging face. How can i do it?

Hi, is cpu inference possible in colab?

Nice work!
My only confusion is how you change the style of the whole image instead of just the face. Usually, StyleGAN generates aligned face images by fine-tuning the FFHQ checkpoint. How does the pix2pix model trained with these face image pairs work with the full image or frame.

Interesting topic... I wonder how you trained the model, especially the augmentation part.
Fixed crop limitation is a well-known problem and would like to know how you handle it. :)

The following are input image, my training output from pair label supervision, and the output from your test model。
I trained my model (Super-Resolution model) on the images from your model outputs, I find it difficult to change the facial features。
Like the eyes and face texture are changed, how to do it ? I use L1Loss (weight is 1) + PerceptualLoss (weight is 1)+ GANLoss (weight is 0.1),

Hi would you share your training code

Was trying the ArcaneGan video colab but I am having a tuple issue can you please help, i am really excited to try the Arcane video can you please help out

Hello, thank you for your wonderful work.

What loss function do you use?

• How to learn to improve performance?

can I use pix2pix HD for img2img translation for this case? thanks.

can i run this project on my laptop?

Hi, how many pictures did you use to train stylegan model?

Hi,

I trained a model with unet_learner but I can't convert it to jit.

I run the following code:
torch.jit.save(torch.jit.script(learn.model), 'jit.pt')

Here is the error:

UnsupportedNodeError: GeneratorExp aren't supported: File "/usr/local/lib/python3.7/dist-packages/fastai/callbacks/hooks.py", line 21 "Applies hook_functomodule, input, output." if self.detach: input = (o.detach() for o in input ) if is_listy(input ) else input.detach() ~ <--- HERE output = (o.detach() for o in output) if is_listy(output) else output.detach() self.stored = self.hook_func(module, input, output)

May I know how you convert it to a jit model?
Thanks

how do you make sure the background and expression similarity between the generated input(face) and target(style face) ? I find that the style is too weak when less finetune and the similarity is too weak when more finetune, how do you solve it ? Would you like to share the paired datasets generated code with me ? thanks a lot ~

Hi,

I want to train the Fastai u-net model. However, when I try to train the critic (learn_critic.fit_one_cycle(6, 1e-3)), I get the following error:

CUDA out of memory. Tried to allocate 4.00 GiB (GPU 0; 14.76 GiB total capacity; 9.78 GiB already allocated; 891.75 MiB free; 12.57 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

The GPU is a Tesla T4 with 16 GB of VRAM. My batch size is 4 and the training images size is 512*512. I also tried with lower numbers, but I'm still getting the same error.

I noticed that you updated the model, but did not find the new version in the code repository or release.

Добрый вечер.В ArcaneGAN на colab for videos,выдаёт ошибку:

RuntimeError: CUDA out of memory. Tried to allocate 2.80 GiB (GPU 0; 11.17 GiB total capacity; 5.74 GiB already allocated; 2.21 GiB free; 8.44 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Помогите пожалуйста!

Is there any way to increase the output resolution on collab, as the current output is quite low.

Thank you!

Colab video : Connection errored out.

I trained a model but the background is blurring. how to improve it?
Thanks for your replying.

It's so cool but I'd like to try to generate the outputs with Japanese anime style.
What can I do that?

Hi, what does the architecture look like?
Is it similar to Pix2Pix?
And for processing of the video, are you doing anything extra to make sure the frames are consistent?

Hi, do you make face alignment before you give it to your Unet model?