search

curryjung / InjectFusion_official

119 stars 3 forks source link

DiffStyle on my own dataset #7

Open Heanhu opened 1 year ago

Heanhu commented 1 year ago

Hi, I trained a diffusion model on P2 weighting using my own dataset (medical images: color fundus images) and then used the code here for style transfer, but the result, as with style, did not change. What parameters can be adjusted, or have you encountered similar problems? Thanks！

curryjung commented 1 year ago

I don't have any domain knowledge about the medical images you used in training, and not sure exactly what the problem is. But, increasing h_gamma might help to improve the content injection for style transfer.

A detailed explanation of your work(what is style image and content image...) will help us to give you better comments:)

Thank you