How to improve the definition of the generated image in a larger image?

[Baenny]

Hello, thanks for the great work! However, I have one problem. When I put the generated frames to the larger original image, the result is low definition. And the result is similar when I use 256 model and 512 model. So is it because that the generator is not super-resolution? If I use the video super-resolution method to process the generated image, is it effective? The boundary of generate image is obvious in the original image.

[AliaksandrSiarohin]

This is a common problem. Super resolution will not be helpful here. You can try to combine the 2 images with soft mask that is not hard near the border, but which is linearly decaying.

[Baenny]

这是一个常见的问题。超分辨率在这里没有帮助。您可以尝试将 2 个图像与在边界附近不硬但线性衰减的软蒙版结合起来。

Thank u! I'll try it.