search

csxmli2016 / textbsr

This is a simple text image blind super-resolution model, using BSRGAN
Other
87 stars 5 forks source link

Does textbsr support Japanese or other languages? #2

Open Ken1256 opened 1 year ago

Ken1256 commented 1 year ago

Does textbsr support Japanese or other languages?

csxmli2016 commented 1 year ago

Does textbsr support Japanese or other languages?

Hi, this model is trained using Chinese and English characters, without any semantic constraints. So if the degradation is not severe, this model can also perform well (see https://imgsli.com/MTg0NjU0). Note that I use cnstd for text region detection, which is also trained for English and Chinese text detection. It is better to use the text detection model according to the language.

Ken1256 commented 1 year ago

Can you add a Japanese text detection to textbsr? If textbsr only changes the text area, you can use text detection after textbsr and the mask will be more accurate.

csxmli2016 commented 1 year ago

Reference in ne

Thanks for your kind suggestion. I am not familiar with Japanese text detection, so this may remain to other users who are interested in transferring this framework to other languages. Generally, if the text detection is not accurate, that means this text region may have lower image quality, which may also fail to restore using our textbsr. I understand your proposed pipeline. This seems not suitable for this simple framework.

Ken1256 commented 1 year ago

I tried EasyOCR for Japanese text detection and the results were impressive, maybe you can add to textbsr. https://github.com/JaidedAI/EasyOCR test01 test02

csxmli2016 commented 1 year ago

I tried EasyOCR for Japanese text detection and the results were impressive, maybe you can add to textbsr. https://github.com/JaidedAI/EasyOCR test01 test02

Wow, it is great! The performance is impressive. I will incorporate it into this framework by the end of this month. Thanks.

skunkwerk commented 10 months ago

This is a great package. I'd like to use this for the Korean language - do you have any training code I could use with a custom dataset? Korean is supported by easyOCR and paddleOCR as well.

csxmli2016 commented 10 months ago

This is a great package. I'd like to use this for the Korean language - do you have any training code I could use with a custom dataset? Korean is supported by easyOCR and paddleOCR as well.

No problem. You can send me the request to my email csxmli@gmail.com.