Best practices for using PaddleOCR in other programming languages ​​and interpreting its logs

[NabiKAZ]

For use in other languages, I try to run the command by terminal or operating system and analyze the output log by RegExp.
For example in nodejs using child_process something like this:

import { spawn } from 'child_process';

 const process = spawn('paddleocr', ['--image_dir', imagePath, '--lang', 'en', '--use_gpu', 'false']);
 process.stdout.on('data', (data) => {
 stdoutData += data.toString();
 });
:
:
 const lines = stdoutData.split('\n');
 const resultLine = lines.find(line => line.includes('INFO: ['));

 if (resultLine) {
 const match = resultLine.match(/\[\[(.*?)\], \('(.*?)',/);
:
:

And also in other languages ​​similar to the same method.

If we ignore the complexities of handling commands and messing with stdout and stderr, etc. Again, doing this has many problems.
For example, we don't always have a fixed output pattern and RegExp always comes with an error.

For example, in these two examples, our text is preceded by (" and (' once:

[2024/08/17 14:14:11] ppocr INFO: [[[207.0, 130.0], [601.0, 130.0], [601.0, 156.0], [207.0, 156.0]], ("salam' khobi' che khabar ", 0.9787972569465637)]

[2024/08/17 14:14:44] ppocr INFO: [[[236.0, 303.0], [626.0, 303.0], [626.0, 334.0], [236.0, 334.0]], ('salam "khobi" che khabar ', 0.9980602860450745)]

Or here where even the found text characters are Escaped!

[2024/08/17 14:15:38] ppocr INFO: [[[204.0, 126.0], [1025.0, 128.0], [1025.0, 160.0], [203.0, 158.0]], ('salam\' khobi\' che khabar salam "khobi" che khabar', 0.9854654669761658)]

Or in cases where there is no text in the photo, like this:

[2024/08/17 13:25:51] ppocr INFO: **********./1.jpg************
[2024/08/17 13:25:52] ppocr DEBUG: dt_boxes num : 0, elapsed : 0.10552430152893066
[2024/08/17 13:25:52] ppocr DEBUG: cls num : 0, elapsed : 0
[2024/08/17 13:25:52] ppocr DEBUG: rec_res num : 0, elapsed : 0.0
Traceback (most recent call last):
 File "<frozen runpy>", line 198, in _run_module_as_main
 File "<frozen runpy>", line 88, in _run_code
 File "C:\Users\Nabi\AppData\Local\Programs\Python\Python311\Scripts\paddleocr.exe\__main__.py", line 7, in <module>
 File "C:\Users\Nabi\AppData\Local\Programs\Python\Python311\Lib\site-packages\paddleocr\paddleocr.py", line 894, in main
 for line in res:
TypeError: 'NoneType' object is not iterable

I don't really know what to set the criteria, is NoneType words or something like dt_boxes num : 0, enough?
But here the very big problem that can happen is that if the text inside my photo really has such words, then what unexpected things will happen!

Such outputs seem to be for direct use in code. But this is a log for me and cannot be interpreted directly.
These were examples, and even if all these cases were resolved, many unforeseen cases would remain.
It is really surprising for me to do such tasks that can be easily solved and but here it wastes a lot of energy.

• But my question is, is this correct in principle?
• Doesn't PaddleOCR offer a more basic method? To use other languages ​​to reach exactly the desired text? Like API or whatever...
• Do you have a better and more accurate idea for me to interpret the log?