Cannot generate images with as high visual quality as given examples in the paper

[whhamber]

Thanks for the released code! I like the proposed method in your paper and tried to apply it in my own task.

When I ran the Diverse Image Generation task, I got unexpected results: the generated output is either messy with low visual quality, or is nearly identical to the input with almost nothing changed.

I tried using smaller "sigma". When sigma is set up to 0.1 or 0.3, the generated output is identical to the input.

I wonder that is there anything wrong with the released code?

[SoTTM]

I have same problem. Did you fix it? it seems no response to this problem from their team.