kaonashi-tyc / rewrite Goto Github PK

Your work is very impressive to me. I think this work may be applied for font style copying among different languages.

But as you said in the front page, I've found some of images that your network generated are blurry. That is one of characteristic of L2(mse) loss.

So I think it would be worth to try GAN. I guess you already know but I just write a brief explanation about it. When you train a GAN, you use and train another network(called discriminator) simultaneously that predicts image whether it is original or generated one.

This is a tensorflow implementation of GAN for Superresolution implemented in Tensorflow.
https://github.com/buriburisuri/SRGAN

While training, the frames go full white for few steps and then after that all steps has black frames.
I passed exactly the same arguments as you explained.
Please tell me how to fix this problem, screenshot of frames is attached.

Hi Taylor,

The project is interesting. I tried to run it on my computer. However I get an error with the commend:

python rewrite.py --mode=train
--model=medium
--source_font=path_to_save_bitmap/src.npy
--target_font=path_to_save_bitmap/trg.npy
--iter=3000
--num_examples=2100
--num_validations=100 --tv=0.0001
--alpha=0.2 --keep_prob=0.9
--num_ckpt=10
--ckpt_dir=path_to_save_checkpoints
--summary_dir=path_to_save_summaries
--frame_dir=path_to_save_frames

Do you know why? Thanks.

BTW, some functions you use are deprecated by tf. Maybe you want to update them.

Hello, we are a company that does intelligent design business, because business development requires the ability to learn deeply in fonts.

I saw you release programming about Chinese font in-depth learning on github, hoping to ask if there is any intention to cooperate and the form of cooperation.

Description of the capabilities we need:

A. Generate a set of Chinese fonts for class on the basis of as few font samples as possible and as little time as possible;

B. The possibility of multiplexing tools for custom font depth learning; look forward to your reply.