taki0112 / stargan-tensorflow Goto Github PK

can this code run Facial expression synthesis results on the RaFD dataset.？

thanks for you sharing your work, Now, My test image is 128*128, the result as follow , is reasonable? THX!

result:

i hasnot rafd dataset， Can you share your rafd data? thk you

Hi, buddy, I got two gpus, can you tell me use multi-gpu training, please? Should I modify this to call the GPU command:
gpu_device = '/gpu:0'
train_dataset = train_dataset.\ ....
to: gpu_device = '/gpu:0'&& '/gpu:1' ?
train_dataset = train_dataset.\ ....
Thanks!

这个可以用来进行表情图像的生成吗？

there should has mask in the origin paper,for different dataset,
but , i can not find it in this code、、、

Excuse me, can you share the Pretrained model?There is no way to click in from the link you gave.It's only for academic purposes.This is my email address [email protected],thank you!

It is miss module data??

Hello, enjoy your work. But the pre-trained model checkpoint for 128x128 does not contain facial expression? it works only for this five attribution 'Black_Hair', 'Blond_Hair', 'Brown_Hair', 'Male' and 'Young' ? @taki0112

thanks for your contribution

but i want to try my dataset, but I donnot know how to create labels, when i train main.py , always get following error :
IndexError: list index out of range

could you please tell how to crrate own dadaset detailed

thanks

I found that there is a difference in the classification weight between it and the pytorch version of the code. What effect does this have?

When I test your pre-trained model, it always raises an error on custom label even the default label

input:
python3 main.py --phase=test --selected_attrs=['Black_Hair','Blond_Hair','Brown_Hair','Male','Young'] --custom_label=[1,0,0,0,0]

error:
main.py: error: argument --custom_label: invalid int value: '[1,0,0,0,0]'

Front page recommends:
python download.py celebA

But download.py will skip the celebA dataset if the celebA directory is present, which it is in the repository.

I could have fixed this by issuing pull request on a version without celebA but I see reasons one might want to include that example directory so I thought I would just bring it up as an issue to be aware of. Right now if somebody clones the project and starts running the example commands this will fail to get the dataset.

sorry,when i use the pre-trained model, i cann't open this file.
the reason is UnicodeDecodeError: 'gbk' codec can't decode byte 0x86 in position 58: illegal multibyte sequence.
And i can't use chardet to get its coding scheme。
So，could you tell me this?（I have used gbk and utf-8）

I am unable to understand how do we define custom labels for other attributes?

hello,thank you for sharing your code.When I reproduced your algorithms and code, it always raises an error, I think there is something wrong that caused by tensorflow version,please help me solve this error or give me some advice,thank you very much.
error:
from tensorflow.contrib.data import prefetch_to_device, shuffle_and_repeat, map_and_batch
ImportError: cannot import name prefetch_to_device
train_dataset = tf.data.Dataset.from_tensor_slices((Image_data_class.train_dataset, Image_data_class.train_dataset_label, Image_data_class.train_dataset_fix_label))
AttributeError: 'module' object has no attribute 'data'

Hi~I meet a problem while i try to use your code to do painting style transfer.
I find that during training, the discriminator loss may be negative cause sometimes the generated image gets a lower loss than the real one from the discriminator which is not sufficiently trained.
Have you ever met the same problem?Will it fo favor to the final performance of our model?
What if I use a pre-trained discriminator?
Waiting for your reply.