solved

Hi I have tried the model, really cool!
Just wondering whether it always generate 64 images? is there a way to output the 'best one'/ 'best ones'?
Or do the 64 output from the style palette have style names if I just want to have one of a few output?

Thank you

I dont understand how am i suppose to generate caricature from input photos.
I downloaded the checkpoint "001000.pt"

When I run
test.py --invert_images --ckpt checkpoint/StyleCariGAN/001000.pt

I get the error: RuntimeError: File samples/samples\jeniffer256.pt cannot be opened.

Am I somehow suppose to generate .pt files for each of my Images.

connect to the link failed,
https:host not found
unlink:no such file or directory
no matched on pattern '120000.pt?dl=0'

Here is colab version if anyone wants to try it:
https://colab.research.google.com/drive/1HDRQGm7pvC9mAb6Lktoft_SmY9sCq_Qg?usp=sharing

Hi this work is fantastic!

I wonder know, is there a way to generate clothes style mapping the cari avatar?

For example, the man wears a suit, our model generates the full body portrait in cari style.

Thanks first!

I meet this mistake when I try your test.py by our images.

I guess if it because we don't have the args.image_name+'.pt' in our images folder.
and it occurs

by the way, our environment is suitable for this project. running on 2080ti + cuda10.1 + torch 1.5.1

Need I train a model myself or where could I get the pt for each images in my folder.

Hello, when I executive "bash download.sh", it appears that www.dropbox.com is not connected.
Could you upload the pretrained weights? Thank you very much!

How to train the attribute classifiers?
Where is the training code？
Thank you.

Firstly, this work is damn fascinating! Admiring you all! Thanks!

I am a beginner of GAN, and I wonder know a naive question, Is InterFaceGAN's face editing is prior than styleCariGAN in paper?

For example, if I understand correctly, I want to draw an open-mouth face caricature from a close-mouth photo, I need to edit face firstly using some GAN like InterFaceGAN, and then use the prior output image, send as input to styleCariGAN?

I test styleCariGAN's xx.pt invert codes seems not compatible with InterFaceGAN's xx.npy format, couldn't edit after style-cari processing directly. haha

please add a google colab for inference

! WARNING !!

Environment

Anaconda environment under Arch Linux.
Python = 3.7 with other requirements with matching versions specified in the README.
(Pytorch=1.3.1)

Steps I took

Run the test.py with

python test.py --ckpt checkpoint/StyleCariGAN/001000.pt --input_dir examples/samples --output_dir examples/results --invert_images

Expectation

The code should run without error

What I get

The code fails to run and I get this error message.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Your compiler (c++) is not compatible with the compiler Pytorch was
built with for this platform, which is g++ on linux. Please
use g++ to to compile your extension. Alternatively, you may
compile PyTorch from source using c++, and then you can also use
c++ to compile your extension.

See https://github.com/pytorch/pytorch/blob/master/CONTRIBUTING.md for help
with compiling PyTorch from source.
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

                          !! WARNING !!

platform=sys.platform))
Traceback (most recent call last):
File "test.py", line 10, in
from invert import *
File "/home/ycjung/Downloads/git/StyleCariGAN/invert.py", line 12, in
from exaggeration_model import StyleCariGAN
File "/home/ycjung/Downloads/git/StyleCariGAN/exaggeration_model.py", line 7, in
from model import PixelNorm, EqualLinear, ConstantInput, StyledConv, ToRGB, ExaggerationLayer
File "/home/ycjung/Downloads/git/StyleCariGAN/model.py", line 13, in
from op import FusedLeakyReLU, fused_leaky_relu, upfirdn2d
File "/home/ycjung/Downloads/git/StyleCariGAN/op/init.py", line 1, in
from .fused_act import FusedLeakyReLU, fused_leaky_relu
File "/home/ycjung/Downloads/git/StyleCariGAN/op/fused_act.py", line 15, in
os.path.join(module_path, "fused_bias_act_kernel.cu"),
File "/home/ycjung/.conda/envs/stylecarigan/lib/python3.7/site-packages/torch/utils/cpp_extension.py", line 661, in load
is_python_module)
File "/home/ycjung/.conda/envs/stylecarigan/lib/python3.7/site-packages/torch/utils/cpp_extension.py", line 841, in _jit_compile
return _import_module_from_library(name, build_directory, is_python_module)
File "/home/ycjung/.conda/envs/stylecarigan/lib/python3.7/site-packages/torch/utils/cpp_extension.py", line 1052, in _import_module_from_library
return imp.load_module(module_name, file, path, description)
File "/home/ycjung/.conda/envs/stylecarigan/lib/python3.7/imp.py", line 242, in load_module
return load_dynamic(name, filename, file)
File "/home/ycjung/.conda/envs/stylecarigan/lib/python3.7/imp.py", line 342, in load_dynamic
return _load(spec)
ImportError: /home/ycjung/.conda/envs/stylecarigan/lib/python3.7/site-packages/torch/../../../libstdc++.so.6: version `GLIBCXX_3.4.29' not found (required by /tmp/torch_extensions/fused/fused.so)

Analysis

The testing code internally invokes g++ and compiles external modules in op/, which contains several cpp and cu files. The invoked g++ is installed on your system and is probably new (In my case it was 11.1.0). The g++ used for compiling PyTorch 1.3.1 in anaconda environment is old and does not produce binaries compatible with the external module compiled with the new g++.

Troubleshooting

It seems libtorch.so in pytorch/pytorch package is compiled with GCC 7.3.1. So install GCC 7.3.0 in your anaconda environment. (7.3.1 is not in anaconda)

conda install -c anaconda gxx_linux-64=7.3.0

Then run the code with environment variables specifying GCC compilers installed in the conda environment

CC=x86_64-conda_cos6-linux-gnu-gcc CXX=x86_64-conda_cos6-linux-gnu-g++ python test.py --ckpt checkpoint/StyleCariGAN/001000.pt --input_dir examples/custom --output_dir examples/results --invert_images

This solution got the code running fine. I hope it does to your environment!

According to the configuration in your paper, how long does it take to train by default? emmmm, can the 24g 3090 be trained?

Hello, file https://www.dropbox.com/s/ryc7v1ghsm54mr8/001000.pt?dl=0 is not available anymore. Please, reupload it.

hi @wonjongg, your work is interesting!
Can you share with me the pre-trained model Caricature StyleGan2 fine-tuned on the caricature dataset (WebCaricature)?
Thank you!

Hello,
Your code need updat here:

• in model.py, line 19:
  old code:
  self.resnet = resnet18(pretrained=True)
  new code:
  self.resnet = resnet18(weights='ResNet18_Weights.DEFAULT')

• in invert.py, line 19:
  old code:
  perceptual = torchvision.models.vgg16(pretrained=True)
  new code:
  perceptual = torchvision.models.vgg16(weights='VGG16_Weights.DEFAULT')

And your code not Windows compatible. Need install Visual Studio Community edition and add PATH this line (where have cl.exe):

• for Windows x64: C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.33.31629\bin\Hostx64\x64
• for Windows x86: C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.33.31629\bin\Hostx64\x86

I then created a testimages folder with an in and an out subfolder. I put in a JPG image and ran the test. This is what I got:
J:\StyleCariGAN-master>test.py --ckpt checkpoint\StyleCariGAN\001000.pt --input_dir testimages\in --output_dir testimages\out --invert_images optimizing w optimizing w: loss_pixel: 0.4483; loss_feature: 4.0835: 100%|█| 250/250 [00:23< optimizing wp optimizing wp: loss_pixel: 0.2393; loss_feature: 2.2942: 100%|█| 2000/2000 [04: Traceback (most recent call last): File "J:\StyleCariGAN-master\test.py", line 120, in <module> invert(g_ema.photo_generator, perceptual, photo, device, args) File "J:\StyleCariGAN-master\invert.py", line 341, in invert torch.save(result, args.result_dir + f'/{args.image_name}.pt') File "C:\Users\Mykee\miniconda3\lib\site-packages\torch\serialization.py", line 376, in save with _open_file_like(f, 'wb') as opened_file: File "C:\Users\Mykee\miniconda3\lib\site-packages\torch\serialization.py", line 230, in _open_file_like return _open_file(name_or_buffer, mode) File "C:\Users\Mykee\miniconda3\lib\site-packages\torch\serialization.py", line 211, in __init__ super(_open_file, self).__init__(open(name, mode)) FileNotFoundError: [Errno 2] No such file or directory: 'testimages\\in/testimages\\in\\20200913_124752.pt'

The .pt file(s) has not been generated or saved, so it cannot be found. How can I fix this?

why I can't download pre-trained model weights