search

ZichengDuan / TheChosenOne

Unofficial implementation of the paper "The Chosen One: Consistent Characters in Text-to-Image Diffusion Models"
https://arxiv.org/abs/2311.10093
237 stars 22 forks source link

Bad results #11

Open sharkDDD opened 8 months ago

sharkDDD commented 8 months ago

I did not modify any code, I simply ran the training and inference programs directly, and the results were very poor. I'm not sure what the problem is. As the loop increases, the results get worse. Is there a problem somewhere? loop=0: photo_man__sitting_on_a_rocket loop=1: photo_man__sitting_on_a_rocket

oguzcana commented 8 months ago

Same here

Marstaos commented 8 months ago

This looks like it's due to the model not converging. Boost your training reps to checkpoint = 2000 try?

oguzcana commented 8 months ago

This looks like it's due to the model not converging. Boost your training reps to checkpoint = 2000 try?

It is supposed to converge much faster (according to paper, as far as I understand) and it self stops saying it has converged, before hitting the max iteration limit (90%) convergence. Were you able to train anything successful?