because of low quality of data set, it can't use to generate high resolution image.

just use it to fine-tune the action in lower weight

maybe under 0.8

in the next version，I will find or create higher data set to train