dmMaze
A selection of bugs and suggestions
Preface: There is the latest commit, which is automatically downloaded from this repo. version: 1.4.0 branch: dev Commit hash: 7aa9218fbdcee82a088212af8323834c050124b2
Bugs
- [ ] 1. The "text detector" problem. Like a complete idiot, I’m used to manually highlighting clouds and setting up recognition. Let's say I recognized everything, now I need to jam the pages. I turn on the cleaner.... But it doesn’t work without a text detector. No problem, we add a text detector and...... it removes all the blocks that I spent 50 minutes arranging. Yes, I could turn on the wedge right away... But....
- Sequence of text export. It so happened that when exporting text, the sequence is always different. For example, in the interface there is block 1, this is block 5, and so on. Because of this, I had to create an exporter from JSON in which for some reason everything is in the right order (how did this happen (irony)).
- [x] ~~3. Change settings when scrolling the mouse wheel. It’s not that I don’t like it, but in the process of flipping through the settings I constantly change the computing core, for example, to HIP, and the program crashes in agony, complaining about not supporting HIP~~
- [ ] 4. Removing a block results in deleting the mask. So this seems like a feature, but for a long time I didn’t understand why the program didn’t always paint over the blocks, but now it’s clear that when you delete a block, the mask is also deleted
https://github.com/dmMaze/BallonsTranslator/assets/57861007/7eac0f44-95f2-4387-93f0-1e7d12b26aab
Offers
-
[ ] 1. OCR engines The most common thing is that these are other OCR engines. Yes, the current ones are designed for manga, they understand its fonts and everything else. The problem is that for other languages (for example English) or the same handwritten Japanese, GoogleLens and for example Paddle OCR do a better job.
-
[x] ~~2. Numbering. I constantly encounter this bug (downloading edits), of course it can be solved stupidly by adding a zero to the numbers, but it would be nice to have some sorting.~~
-
[ ] 3. Folder with subfolders. The idea is stupid, but what if we make a hierarchy of folders. Suppose we provide a folder with project folders as input. We pass it to the program, and it indexes the contents, and then, having selected the folder with the title, it creates all the directories and a JSON project file in it. The folders in which we worked are highlighted with an asterisk.
-
[ ] 4. Translation of individual files. Let's say we throw a file (picture) into the program. The translator generates a temporary folder (or in its program folder), saves the result in it, and clicking on save file deletes all temporary files.
-
[ ] 5. Fix 1 bug. I think we can do the fix as stupidly as possible by saving everything to a file and then exporting it. For settings, when selecting inpaint, we automatically check the text detector setting, because if you run only inpaint, it will not do anything.
-
[ ] 6. Start recognition one page at a time. Well, everything is clear here, I would like to have the “start translation” function only for the current page.
-
[ ] 7. Reworking the system for working with text. The problem is, let’s say I have ready-made clouds and a ready-made text preset. I apply the preset to all blocks.... And zero results.... That is, I would like to just have a button to apply the “preset” to all. Maybe it exists, but I'm stupid and don't understand how.....
Really lacks the ability to select a different OCR, for English texts for example. I have api key deepl and the translation became much better, but why the text is moved to a new line, when it could remain one sentence in the editing blocks, and in the blocks of text on the image itself it is automatically moved when it rests on the edge of the square.
And of course I would like to be able to save everything as a PSD file for further processing in Photoshop. But I guess python is not capable of this....
And of course I would like to be able to save everything as a PSD file for further processing in Photoshop. But I guess python is not capable of this....
PSD is a proprietary format. There are libraries that can work with it, but the current architecture does not allow you to just take it and translate it into psd. I started translating to PSD myself and was able to separate the layers so far. Text export is not working yet