Hi, thanks a lot for this awesome work. I have few questions about your editing settings.

1. How did the editing coefficients are chosen(-40~40)?
2. Why the L2 norms of your attached editing hyperplane are not 1.0?
   As I tested e2style, I found the distribution of encoded latents is very different from psp's latents, so I guess the editing coefficients might need to change somehow. However, I don't find any details in the paper.
   May you please help me out with the details I have missed?

How do you perform the face swapping with the code?

F:\ProgramData\Anaconda3\python.exe F:/work/mingxingshipin/opencv_Real_Time_Image_Animation-master/e2style-main/scripts/inference.py
Inference: Results are from Stage1.
Loading E2Style from checkpoint: F:/work/mingxingshipin/opencv_Real_Time_Image_Animation-master/e2style-main/pretrained_models/sr.pt
Loading dataset for celebs_super_resolution
Performing down-sampling with factors: [1, 2, 4, 8, 16, 32]
0%| | 0/4 [00:25<?, ?it/s]
Traceback (most recent call last):
File "F:/work/mingxingshipin/opencv_Real_Time_Image_Animation-master/e2style-main/scripts/inference.py", line 120, in
run()
File "F:/work/mingxingshipin/opencv_Real_Time_Image_Animation-master/e2style-main/scripts/inference.py", line 77, in run
result_batch, latent_batch = net(input_cuda, randomize_noise=False, resize=opts.resize_outputs, return_latents=True)
File "F:\ProgramData\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "..\models\e2style.py", line 100, in forward
return images, result_latent
UnboundLocalError: local variable 'images' referenced before assignment

Super resolution got bad results, can you give some advice? The original image is just a problem with the mouth. After super-resolution, it has some male characteristics.

Actually only the mouth area needs to be optimized

What a wonderful job！ Could auther sharing the enviroment of this project? Thanks!

Why is the gloss of the face surface of the image generated by my training insufficient

I've just read through your paper and it's superb.
I created this issue just to say that I'm looking forward to your code.
Hope you can release the code soon!

still bad inversion but i dont know why

Great work, here I got a question about how to train the 2-3th stage on the application of superresolution and so on ? It seems it can only can train the multi stage on GAN inversion

How to Crop out the center area of Is and paste it onto It?

How do you perform the face swapping with the code?

Great work！How to provide test images for inpainting

Hello, Im trying to do some images inversions that contain earrings, nose piercings that kind of stuff and I notice that are they are not captured in the inverted image, e.g.:

Any tips to make the algorithm more sensitive to local features? Thank you!

How to get the editing directions from this repository?

why above picture of donald trump inversion is bad and another is good ??