Finetuning
#38
by
juliajoanna
- opened
Hi,
great work!
I am currently finetuning Zeroscope on my own dataset using the code from Github. The problem is that the finetuned model produces a lot of noise and artifacts on generated videos. I wonder what the reason is. How did you finetune Modelscope? What hyperparameters, like fps, dataset normalization or frame step, did you use? Did you train both on videos and images or not?
Thanks in advance.