dmMaze / BallonsTranslator

深度学习辅助漫画翻译工具, 支持一键机翻和简单的图像/文本编辑 | Yet another computer-aided comic/manga translation tool powered by deeplearning
GNU General Public License v3.0
2.48k stars 172 forks source link

Great request, can you let the translation model combine the recognized words when translating so that the AI ​​model can identify and translate based on the context? #533

Open asdads123 opened 1 month ago

asdads123 commented 1 month ago

The author, can you ask the translation model to combine the recognized words when translating and let the AI ​​model contact the context for recognition and translation? The quality of the AI ​​model's single-word recognition and translation is hard to describe. I used gemini1.5pro to input the context of the text into an AI translation one by one. The effect was surprisingly good. If the recognized text can be combined, the AI ​​model can contact the context for recognition and translation. I think even the 1.8B and 7B model translation quality will be pretty good. My prompt is: You will play a professional translator and translate all my next words into Chinese. It needs to be honest, elegant and emotional. This is a line from a youth comic. You should be more suitable. . 作者大大,可以让翻译模型翻译的时候,将识别的文字和起来让AI模型联系上下文进行识别翻译吗?AI模型单字识别,翻译真是质量一言难尽,我用了gemini1.5pro将文字上下文意一一输个AI翻译,效果意外的好,如果能实现将识别的文字和起来让AI模型联系上下文进行识别翻译。我想就算是1.8B和7B模型翻译的质量也会很不错。我的提示语是:你将扮演一名专业的翻译家,将我接下来的话全部翻译成中文,要求信达雅,富有感情,这是一部青春少年漫的台词,你应该更贴合一些。

asdads123 commented 1 month ago

By the author, what I mean is to combine the recognized words and give them to AI, let AI identify the characters and the style that should be translated, and then translate it through the context. May I? I beg you greatly! It’s so tiring to lose to AI one by one (crying) 作者大大,我的意思是将识别的文字合起来,交给ai,让AI识别出人物和应该翻译出的风格,再通过上下文进行翻译。可以吗?大大跪求!一一输给AI实在是太累了(哭)

dmMaze commented 1 month ago

现在确实有技术可以给整本漫画角色聚类并推断对话框关系,不过角色身份和 prompt 肯定还得用户打一次... 我下半年有空弄一下

asdads123 commented 1 month ago

现在确实有技术可以给整本漫画角色聚类并推断对话框关系,不过角色身份和 prompt 肯定还得用户打一次...我下半年有空弄一下 谢谢大大,现在大大模型例如gemini,如果给出通过角色的上下发言并用,隔开再叫个AI翻译,其质量是非常高的,但如果只是单纯的调用模型在翻译上都基本是机翻(哭) ![Uploading 屏幕截图 2024-08-08 200850.png…]()

asdads123 commented 1 month ago

屏幕截图 2024-08-08 200850

asdads123 commented 1 month ago

屏幕截图 2024-08-08 200900 屏幕截图 2024-08-08 200914 屏幕截图 2024-08-08 200926

asdads123 commented 1 month ago

大大这是我调用Google模型时的图片

asdads123 commented 1 month ago

还有大大能增加个手动添加ocr识别框的功能吗?(please)在翻译过程容易出现有原文没能被识别的情况,如果能手动添加ocr框,就能够手动右键ocr进行翻译