pacifinapacific / stylegan_latenteditor Goto Github PK

What is the implementation of style transfer, I want to run a test.Thanks！

Hi, thanks for your repo.
I have downloaded the official TensorFlow model from https://drive.google.com/uc?id=1MEGjdvVpUsu1jB4zrXZN7Y4kBBOzizDQ.
Then I successfully run weight_convert.py.

However, when I am trying to run the code, there are errors when load the model:

`Missing key(s) in state_dict: "g_mapping.dense0.weight", "g_mapping.dense0.bias", "g_mapping.dense1.weight", "g_mapping.dense1.bias", "g_mapping.dense2.weight", "g_mapping.dense2.bias", "g_mapping.dense3.weight", "g_mapping.dense3.bias", "g_mapping.dense4.weight", "g_mapping.dense4.bias", "g_mapping.dense5.weight", "g_mapping.dense5.bias", "g_mapping.dense6.weight", "g_mapping.dense6.bias", "g_mapping.dense7.weight", "g_mapping.dense7.bias", "g_synthesis.torgb.weight", "g_synthesis.torgb.bias", "g_synthesis.blocks.4x4.const", "g_synthesis.blocks.4x4.bias", "g_synthesis.blocks.4x4.epi1.top_epi.noise.weight", "g_synthesis.blocks.4x4.epi1.style_mod.lin.weight", "g_synthesis.blocks.4x4.epi1.style_mod.lin.bias", "g_synthesis.blocks.4x4.conv.weight", "g_synthesis.blocks.4x4.conv.bias", "g_synthesis.blocks.4x4.epi2.top_epi.noise.weight", "g_synthesis.blocks.4x4.epi2.style_mod.lin.weight", "g_synthesis.blocks.4x4.epi2.style_mod.lin.bias", "g_synthesis.blocks.8x8.conv0_up.weight", "g_synthesis.blocks.8x8.conv0_up.bias", "g_synthesis.blocks.8x8.conv0_up.intermediate.kernel", "g_synthesis.blocks.8x8.epi1.top_epi.noise.weight", "g_synthesis.blocks.8x8.epi1.style_mod.lin.weight", "g_synthesis.blocks.8x8.epi1.style_mod.lin.bias", "g_synthesis.blocks.8x8.conv1.weight", "g_synthesis.blocks.8x8.conv1.bias", "g_synthesis.blocks.8x8.epi2.top_epi.noise.weight", "g_synthesis.blocks.8x8.epi2.style_mod.lin.weight", "g_synthesis.blocks.8x8.epi2.style_mod.lin.bias", "g_synthesis.blocks.16x16.conv0_up.weight", "g_synthesis.blocks.16x16.conv0_up.bias", "g_synthesis.blocks.16x16.conv0_up.intermediate.kernel", "g_synthesis.blocks.16x16.epi1.top_epi.noise.weight", "g_synthesis.blocks.16x16.epi1.style_mod.lin.weight", "g_synthesis.blocks.16x16.epi1.style_mod.lin.bias", "g_synthesis.blocks.16x16.conv1.weight", "g_synthesis.blocks.16x16.conv1.bias", "g_synthesis.blocks.16x16.epi2.top_epi.noise.weight", "g_synthesis.blocks.16x16.epi2.style_mod.lin.weight", "g_synthesis.blocks.16x16.epi2.style_mod.lin.bias", "g_synthesis.blocks.32x32.conv0_up.weight", "g_synthesis.blocks.32x32.conv0_up.bias", "g_synthesis.blocks.32x32.conv0_up.intermediate.kernel", "g_synthesis.blocks.32x32.epi1.top_epi.noise.weight", "g_synthesis.blocks.32x32.epi1.style_mod.lin.weight", "g_synthesis.blocks.32x32.epi1.style_mod.lin.bias", "g_synthesis.blocks.32x32.conv1.weight", "g_synthesis.blocks.32x32.conv1.bias", "g_synthesis.blocks.32x32.epi2.top_epi.noise.weight", "g_synthesis.blocks.32x32.epi2.style_mod.lin.weight", "g_synthesis.blocks.32x32.epi2.style_mod.lin.bias", "g_synthesis.blocks.64x64.conv0_up.weight", "g_synthesis.blocks.64x64.conv0_up.bias", "g_synthesis.blocks.64x64.conv0_up.intermediate.kernel", "g_synthesis.blocks.64x64.epi1.top_epi.noise.weight", "g_synthesis.blocks.64x64.epi1.style_mod.lin.weight", "g_synthesis.blocks.64x64.epi1.style_mod.lin.bias", "g_synthesis.blocks.64x64.conv1.weight", "g_synthesis.blocks.64x64.conv1.bias", "g_synthesis.blocks.64x64.epi2.top_epi.noise.weight", "g_synthesis.blocks.64x64.epi2.style_mod.lin.weight", "g_synthesis.blocks.64x64.epi2.style_mod.lin.bias", "g_synthesis.blocks.128x128.conv0_up.weight", "g_synthesis.blocks.128x128.conv0_up.bias", "g_synthesis.blocks.128x128.conv0_up.intermediate.kernel", "g_synthesis.blocks.128x128.epi1.top_epi.noise.weight", "g_synthesis.blocks.128x128.epi1.style_mod.lin.weight", "g_synthesis.blocks.128x128.epi1.style_mod.lin.bias", "g_synthesis.blocks.128x128.conv1.weight", "g_synthesis.blocks.128x128.conv1.bias", "g_synthesis.blocks.128x128.epi2.top_epi.noise.weight", "g_synthesis.blocks.128x128.epi2.style_mod.lin.weight", "g_synthesis.blocks.128x128.epi2.style_mod.lin.bias", "g_synthesis.blocks.256x256.conv0_up.weight", "g_synthesis.blocks.256x256.conv0_up.bias", "g_synthesis.blocks.256x256.conv0_up.intermediate.kernel", "g_synthesis.blocks.256x256.epi1.top_epi.noise.weight", "g_synthesis.blocks.256x256.epi1.style_mod.lin.weight", "g_synthesis.blocks.256x256.epi1.style_mod.lin.bias", "g_synthesis.blocks.256x256.conv1.weight", "g_synthesis.blocks.256x256.conv1.bias", "g_synthesis.blocks.256x256.epi2.top_epi.noise.weight", "g_synthesis.blocks.256x256.epi2.style_mod.lin.weight", "g_synthesis.blocks.256x256.epi2.style_mod.lin.bias", "g_synthesis.blocks.512x512.conv0_up.weight", "g_synthesis.blocks.512x512.conv0_up.bias", "g_synthesis.blocks.512x512.conv0_up.intermediate.kernel", "g_synthesis.blocks.512x512.epi1.top_epi.noise.weight", "g_synthesis.blocks.512x512.epi1.style_mod.lin.weight", "g_synthesis.blocks.512x512.epi1.style_mod.lin.bias", "g_synthesis.blocks.512x512.conv1.weight", "g_synthesis.blocks.512x512.conv1.bias", "g_synthesis.blocks.512x512.epi2.top_epi.noise.weight", "g_synthesis.blocks.512x512.epi2.style_mod.lin.weight", "g_synthesis.blocks.512x512.epi2.style_mod.lin.bias", "g_synthesis.blocks.1024x1024.conv0_up.weight", "g_synthesis.blocks.1024x1024.conv0_up.bias", "g_synthesis.blocks.1024x1024.conv0_up.intermediate.kernel", "g_synthesis.blocks.1024x1024.epi1.top_epi.noise.weight", "g_synthesis.blocks.1024x1024.epi1.style_mod.lin.weight", "g_synthesis.blocks.1024x1024.epi1.style_mod.lin.bias", "g_synthesis.blocks.1024x1024.conv1.weight", "g_synthesis.blocks.1024x1024.conv1.bias", "g_synthesis.blocks.1024x1024.epi2.top_epi.noise.weight", "g_synthesis.blocks.1024x1024.epi2.style_mod.lin.weight", "g_synthesis.blocks.1024x1024.epi2.style_mod.lin.bias".
Unexpected key(s) in state_dict: "fromrgb.weight", "fromrgb.bias", "1024x1024.conv0.weight", "1024x1024.conv0.bias", "1024x1024.blur.kernel", "1024x1024.conv1_down.weight", "1024x1024.conv1_down.bias", "1024x1024.conv1_down.downscale.blur.kernel", "512x512.conv0.weight", "512x512.conv0.bias", "512x512.blur.kernel", "512x512.conv1_down.weight", "512x512.conv1_down.bias", "512x512.conv1_down.downscale.blur.kernel", "256x256.conv0.weight", "256x256.conv0.bias", "256x256.blur.kernel", "256x256.conv1_down.weight", "256x256.conv1_down.bias", "256x256.conv1_down.downscale.blur.kernel", "128x128.conv0.weight", "128x128.conv0.bias", "128x128.blur.kernel", "128x128.conv1_down.weight", "128x128.conv1_down.bias", "128x128.conv1_down.downscale.blur.kernel", "64x64.conv0.weight", "64x64.conv0.bias", "64x64.blur.kernel", "64x64.conv1_down.weight", "64x64.conv1_down.bias", "64x64.conv1_down.downscale.blur.kernel", "32x32.conv0.weight", "32x32.conv0.bias", "32x32.blur.kernel", "32x32.conv1_down.weight", "32x32.conv1_down.bias", "32x32.conv1_down.downscale.blur.kernel", "16x16.conv0.weight", "16x16.conv0.bias", "16x16.blur.kernel", "16x16.conv1_down.weight", "16x16.conv1_down.bias", "16x16.conv1_down.downscale.blur.kernel", "8x8.conv0.weight", "8x8.conv0.bias", "8x8.blur.kernel", "8x8.conv1_down.weight", "8x8.conv1_down.bias", "8x8.conv1_down.downscale.blur.kernel", "4x4.conv.weight", "4x4.conv.bias", "4x4.dense0.weight", "4x4.dense0.bias", "4x4.dense1.weight", "4x4.dense1.bias".

`
I wonder there might be any mismatch between the "karras2019stylegan-ffhq-1024x1024.pkl" I downloaded from the google link and the expected model.

I want to know whether facial_exchange.py corresponds to style_transfer or expression transfer in the paper, because I think the code is similar to style_transfer, but in this case, which one corresponds to expression transfer? Hope you answer your doubts.Thanks!

I am confused about the operation on the synth_img. Why not just pass the synth_img to calculate the loss? Can anyone tell me ?

Hello,What is the implementation of style transfer, is it the same as the file ' facial.exchange.py '?I don't know,can you tell me?

Thank you very much!

Can anyone share weight files? The link provided by the author cannot be downloaded at this time.

Hi, thank you for your good work, I have a question, In paper image2styleGAN++, the author mentioned that they both optimize w and n(noise) , but in your code, I only find w, and find nothing about noise optimize process

Can you please let me know how to change default image? (0.png)

Hello,when I run this .py file,error happend,can you tell me how to solve?thank you!

"RuntimeError: No default TensorFlow session found. Please call dnnlib.tflib.init_tf()."

the complete code in dnnlib/tflib/tfutil.py:
def assert_tf_initialized():
"""Check that TensorFlow session has been initialized."""
if tf.get_default_session() is None:
raise RuntimeError("No default TensorFlow session found. Please call dnnlib.tflib.init_tf().")

There is an application called Style Transfer in the paper “Image2StyleGAN: How to Embed Images Into the StyleGAN Latent Space?” Did anyone try to accomplish it? My result is bad. I have try the loss function include perceptual loss +MSE loss (the paper method) and only the style loss from conv4_2 layer of VGG-16 as well as style loss from conv4_2 layer of VGG_16 +MSE loss.

What's the version of pytorch and tensorflow ? And python ?

I have a tensorflow 2 installed in my machine. So changed the tensorflow imports to the following-
"import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()"

When I try to run weight_convert.py, I get this error shown in the screengrab

I tried but failed to locate the G_synthesis function. Can you help me out. Thanks!

Can you tell me whice ".py" file do the style transfer ?
Thanks you!

I try to use my own image like a cat which resolution is 512x512. I didn't run the aline_images command because I don't want my orignial image to be resized and the human face detector cannot detect the cat face too. But when I set the command args.resolution=512, g_all.load_state_dict() get wrong. The Pytorch weight is still the same as "./karras2019stylegan-ffhq-1024x1024.pt" when i run on resolution 1024x1024.

Below is the wrong information.

python encode_image.py --src_im sample.png --iteration 10 --resolution 512
Traceback (most recent call last):
File "encode_image.py", line 123, in
main()
File "encode_image.py", line 37, in main
g_all.load_state_dict(torch.load(args.weight_file, map_location=device))
File "/home/beryl/anaconda3/envs/py36/lib/python3.6/site-packages/torch/nn/modules/module.py", line 839, in load_state_dict
self.class.name, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for Sequential:
Unexpected key(s) in state_dict: "g_synthesis.blocks.1024x1024.conv0_up.weight", "g_synthesis.blocks.1024x1024.conv0_up.bias", "g_synthesis.blocks.1024x1024.conv0_up.intermediate.kernel", "g_synthesis.blocks.1024x1024.epi1.top_epi.noise.weight", "g_synthesis.blocks.1024x1024.epi1.style_mod.lin.weight", "g_synthesis.blocks.1024x1024.epi1.style_mod.lin.bias", "g_synthesis.blocks.1024x1024.conv1.weight", "g_synthesis.blocks.1024x1024.conv1.bias", "g_synthesis.blocks.1024x1024.epi2.top_epi.noise.weight", "g_synthesis.blocks.1024x1024.epi2.style_mod.lin.weight", "g_synthesis.blocks.1024x1024.epi2.style_mod.lin.bias".
size mismatch for g_synthesis.torgb.weight: copying a param with shape torch.Size([3, 16, 1, 1]) from checkpoint, the shape in current model is torch.Size([3, 32, 1, 1]).

In the paper, in the style transfer section, it is mentioned that the calculation uses crossover operation. Is this roughly the same as the crossover.py in your project file?
If not, what is the correct operation?

Hello. Thank you for a cool work! With latest pytorch I have a message on running facial_exchange.py about variable modification at style_loss optimization backward (line 95).

May be codes need minor fix:

• move synth_img_a=g_synthesis(dlatent_a) and synth_img_a= (synth_img_a + 1.0) / 2.0 below to loss_1.backward().
• Also maybe code requires optimizer.zero_grad() after loss1 optimizer step.

hello! I found the directory and corresponding boundary files is missing in this Github. Will you provided for us?

I am trying to convert my trainned model from tensorflow to pytorch version using weight_convert.py however it shows below error

Traceback (most recent call last): File "weight_convert.py", line 37, in <module> weights =pickle.load(open(tensorflow_dir+weight_name+".pkl",'rb')) File "/home/ahmed/Documents/3DAI/StyleGAN_LatentEditor/dnnlib/tflib/network.py", line 279, in __setstate__ assert state["version"] in [2, 3] AssertionError
thank you in advance for reply