lksshw / SRNet

A pytorch implementation of the SRNet architecture from the paper Editing text in the wild (Liang Wu et al.)
152 stars 35 forks source link

About trainning #35

Open mmmmmjaai opened 11 months ago

mmmmmjaai commented 11 months ago

hi iam beginner, i have some questions about trainning iam trying to train my own model using chinese, but the result is so bad that the style font did not erase and the target font did not gen on o_f.jpg. is i input too many kind of chinese fonts during trainning, or the input bg_image too complex(that i found some input are hard to figure out the detail) i used 50K image as input, also train 50K iteration