Trouble on larger masks?

[hukkelas]

Hi,

Thanks for publishing this source code! It makes testing your model easy :)

I'm currently testing the model on center-crop masks; however, I find that the model is struggling with larger center crops?
The images I tested were 512x512, with a center mask of 256x256.

Here is an example image that I tested, where I got strange corruption in the final result.
Ground truth image:

Input image:

Generated image:

Do you know what might be the cause of this? Or are center-crops difficult for the model?

If you would like to reproduce the result, here is the original mask (256x256 center crop):

If the images are compressed, I can give you a link to the raw images :)

[Ehsan-Yaghoubi]

Hi @hukkelas and @duxingren14

I experienced the same thing. I mean, when the size of the mask is large enough (in comparison with the size of the image), the model reconstructs the image with some artifacts. Did you manage to improve the results?

[hukkelas]

Hi! Sadly I haven't had time to look any further into it, but I experienced this weird behavior over several samples with a centered mask.
Often this "gray pattern" appears in the generated image, but I have no explanation for it. It almost looks like some clipping artifacts?

If you find a solution to it, please update me!

[duxingren14]

I saw similar gray patterns in our test results before. It happens when the mask is too big. My suggestion is to reduce the percentage of hole region by image padding.

[kelisiya]

I saw similar gray patterns in our test results before. It happens when the mask is too big. My suggestion is to reduce the percentage of hole region by image padding.

In addition, there are two bad situations. When the image is very large, the repair will have a fuzzy ground glass effect. When the mask is at the extreme edge, the repair will have strange colors.