search

dailenson / SDT

This repository is the official implementation of Disentangling Writer and Character Styles for Handwriting Generation (CVPR23).
MIT License
956 stars 81 forks source link

Some questions about the generated results #79

Closed jackwolfey closed 2 months ago

jackwolfey commented 3 months ago

I have been succesfully execute your code by using your well trained model in the readme, but the chinese generated results seems always be one particular character style.I use windows paint to write 15 chars, and make a image binaryzation before inference and still have this problem.Is there any way to fix this issue or any way to imporve the generated results? Nice work by the way.

dailenson commented 3 months ago

Hi, there. You‘d better provide more problem details, such as your style samples, your generated results, and which file you execute, test.py or user_generate.py.

jackwolfey commented 3 months ago

Okay, i followed #43 and execute the user_generate.py. Here are my handwriting samples: explorer_VCT9kjGItb Here are my generated results: explorer_veWceKbvN9

dailenson commented 2 months ago

Based on your generated results, we recommend changing the style sample collection approach to enhance performance, as done in stylus-written or finger-written.