-
Many months ago I think it was mentioned that a better high-res model is on the working and will released soon.
It's been a while, and since it's very hard for newbies to understand or follow the c…
-
For example, 2x inference time?
Thanks!
-
Thanks for your sharing!
I tried your model and find something wrong. Here are the details:
**1.** when loading wav2lip.pth, "Missing key(s) in state_dict" occured. I made some changes in infere…
-
Hi,
I am trying to train the expert lip sync discriminator using LRS2 and using the code from the repo as is (without any changes), but I am unsure why is the model converging much quicker than it …
-
Thanks for your dedication and great results, I would like to contribute some of my improvements on Loss Function, Activation, Size of Input Image. I have added some layers and the model can be traine…
ghost updated
2 years ago
-
我租了一台8个GPU、单GPU 84G显存的tesla 100,使用缺省头像生成是没问题的。如果使用自动定义头像(我自己的照片,7MB),创建虚拟角色的步骤没有问题(就是生成这个me.mp4),但在生成demo video的时候失败了,报CUDNN_STATUS_NOT_SUPPORTED
# python general_demo.py --human ./file/input/m…
-
Modify the file hparam
img_size=96,increase ?264?
fps=25,
Where is the neural network architecture modified ?
-
-
运行报错,这是什么问题,paddle.nn.utils里确实没有spectral_norm
Traceback (most recent call last):
File "create_virtual_human.py", line 1, in
from PaddleTools.GAN import FOM
File "E:\GitHub\PaddleBoBo\Padd…
-
I trained the lip_sync expert discriminator (**train_loss: 0.25-0.26 val_loss=0.280**)
Then I used that discriminator to train the wav2lip model without the GAN setup
I reached up to **L1_loss: trai…