Use DG-Font to generate a lots of new images with a few images?

[dagongji10]

@ecnuycxie
If I use 400 fonts to train DG-Font. In the inference stage, I write 20 character images and save them as 80*80 .png. Can I input the 20 images as style-image, and generate other 3735 Chinese characters with my writting style? How can I do this?

[ecnuycxie]

This is a few-shot generation problem. You can use a pre-trained model (using 400 fonts) to achieve it. Specifically, you can create a new folder named "id_401" to store your handwriting style images.

[dagongji10]

This is a few-shot generation problem. You can use a pre-trained model (using 400 fonts) to achieve it. Specifically, you can create a new folder named "id_401" to store your handwriting style images.

OK, I will have a try, Thanks!

[dagongji10]

@ecnuycxie
As you said, I can generate new images now. But the result seems not always fine.

I use Song as the content image, and write chinese image as style image. Sometimes the stroke will be drop.
How to solve this problem?

[ecnuycxie]

Try to use KAI as source font.

[shujian2]

@ecnuycxie 正如你所说，我现在可以生成新图像。但结果似乎并不总是好的。 我用宋作为内容形象，把中文形象写成风格形象。有时中风会下降。 如何解决这个问题呢？

Hello, how is this generated

[jingmingtao]

If I use 400 fonts to train DG-Font. In the inference stage, I write 20 character images and save them as 80*80 .png. Can I input the 20 images as style-image, and generate other 3735 Chinese characters with my writting style? How can I do this?

Hello,I use 400 fonts to train DG-Font. In the inference stage, I write 20 character images and save them as 80*80 .png. Can I input the 20 images as style-image, and generate other 3735 Chinese characters with my writting style? How can I do this?
Can you help me?

[666pz]

Hello, how do you generate personal style font from song style in this picture!!! very grateful for you!!

[TomSuen]

@ecnuycxie As you said, I can generate new images now. But the result seems not always fine. I use Song as the content image, and write chinese image as style image. Sometimes the stroke will be drop. How to solve this problem?

你好，我觉得你训练的结果很不错，能够看出手写字体的不规整特征（而非印刷体）；我这边训练的问题是，生成的字体总是很像某种印刷体或者很规整的钢笔字。请问你的训练集包含哪些样式的字体呢？