jgkwak95 / surf-gan Goto Github PK

In practice, replicating source code is much slower than this article suggests.

My practical operations are as follows:
Download img_align_celeba.zip dataset and unzip it in the specified catalogue.
Run the following code "python train_surf.py --output_dir third --curriculum CelebA"
The results I got are as follows.

The code shows that completing this program needs about 813 hours and I repeated this procedures, only to get the same results.

Hi @jgkwak95,
thanks a lot for sharing the code for your work. When I was learning your project code, I didn't find a place to input my own pictures.

Thank you very much !

Hi @jgkwak95,
thanks a lot for sharing the code for your work, it is very interesting. Can we use a 2D profile face to acquire a frontal face by this work?

Thank you very much !

Hello, thanks for your work, and for making your code available.
I am having issues reproducing your results on CelebA, the network seems to converge up to around iteration 15k, but then the image quality degrades until becoming nothing recognizable.
I used the command python train_surf.py --output_dir exp_name --curriculum CelebA_single, with nothing changed from your code.
The python env was created using your instructions in "Create virtual environment".
I ran the training process 3 times, and consistently obtained this result.
Here are random samples at 15k, and 40k:

Thanks for your great work. Could you tell me how to realize 3d controllable face editing via the code in your project? Thanks a lot.

Hi, thank you for your work, and sharing your code and model.
Unfortunately, the google drive link to your pretrained model seems dead right now, it gives a "Not found" error.

Hi @jgkwak95,
thanks a lot for sharing the code for your work, it is very interesting, the results look impressive. Any chance you may also share the pretrained models? It would be very useful.

Thank you very much !

Thanks for your great work, but I can't find the code about "Explicit control over pose with StyleGAN" from the repo, where are they?

Hi, nice work! I notice that FIDs of Surf-gan look great. Could you provide the pre-trained model which is trained on FFHQ@256?

hello, I set the right data path in curriculums.py and run train_surf.py,but i always get the TypeError: init() missing 1 required positional argument: 'dataset_path' when running up to "fid_evaluation.setup_evaluation"