barisgecer / tbgan Goto Github PK

Hello, thanks for your work!

I have a question about the outputs of the pre-trained model. We acquired the pre-trained model from this link. However, our outputs are not the same as those presented in the paper. Unlike the presented results, our results don't generate neck parts of the 3D faces.

Is there any additional processing part to add neck parts to our results?

I am attaching one of our results and a result from the paper.

Thank you.

Hi，when I run python test.py, it has problem like following picture. How to solve this? @barisgecer

Hello, thanks for your work!

I'm doing experiments for my graduation project, and I'd like to make sure they're consistent with your research. Do you have input vectors or seed values for the faces on your paper?

Thank you.

Hi, I'm following your work! And I wonder if your network adopts some training strategies like ProGAN does, such as progressive growth.

Hi, thank you for your excellent work!
I have acquired the pre-trained model from this link. However, when I run python test.py, the output image is quite strange (shown as follow) and I couldn't get a 3D face.

I have followed #6 and the modification of test.py is as follow:

if __name__ == "__main__":
    misc.init_output_logging()
    np.random.seed(config_test.random_seed)
    print('Initializing TensorFlow...')
    os.environ.update(config_test.env)
    tfutil.init_tf(config_test.tf_config)
    print('Running %s()...' % config_test.train['func'])
    # tfutil.call_func_by_name(**config_test.train)
    from util_scripts import generate_interpolation_images
    config_test.num_gpus = 1
    config_test.desc = 'interpolation-images-' + str(config_test.train.run_id)
    generate_interpolation_images(run_id=32, grid_size=[1,1], duration_sec=5.0, smoothing_sec=0.1)
    print('Exiting...')

#----------------------------------------------------------------------------

I don't know what's wrong with my code and how to fix this problem. Can you help me?

Hello, thank you for releasing the code for this amazing research. Do you plan or releasing the weights from your training?

Hi, thanks for your great job!

However, I am a little confused about the identity-generic maps. As mentioned in the paper, the detailed normal map for skin pores, one of the generic maps, should be added on the identity-specific map for improved rendering. How can we get this detailed normal map? Is there any ground truth for the skin pores?

Looking forward for your reply, thanks a lot!

Hi! I would love to try this out， When do you target releasing the code?

Hello! Thanks for sharing the code and your work.

Can you please point out the code to cylindrically unwrap the mean face for creating a UV representation corresponding to a specific mesh topology.

Thanks!

Hi, I have a question about normal UVmap after reading your paper. The vertices of face can obtain directly from shape UVmap, so the normal UVmap is for recalculating the vertice coordinates to improve the face mesh precision or it is only used to render face obj?

While following the steps mentioned in the readme.md file, I encountered an error for preparing the dataset. create_from_pkl_img_norm function from dataset_tool.py requires some .pkl files and texture,shape,normals which are not provided in the code. I have downloaded the FFHQ dataset and the tfrecords.