search

ankanbhunia / PIDM

Person Image Synthesis via Denoising Diffusion Model (CVPR 2023)
https://ankanbhunia.github.io/PIDM
MIT License
481 stars 62 forks source link

About the color difference between the generated image and ground truth #42

Open murphytju opened 1 year ago

murphytju commented 1 year ago

I trained this generative model with my own data, using the same MSEloss and vbloss as in the code you provided, and after 100 generations of training the network can correctly generate the content of the image, but the color of the image deviates a lot, and I'd like to ask if you've encountered a similar problem.

DoubleCake commented 1 year ago

i got the same problem，but i didnot solve it