Derive the model names from directory names if provided

[timminator]

PaddleOCR 3.0.0/3.0.1 requires specifying the name in addition to the model directory which is not really necessary because they can be derived from the directory name. This improves the user experience by making it easier to use.

If this gets approved, I would appreciate it to be also cherry-picked for the 3.0.2 release.

Fixes #15707

[paddle-bot]

Thanks for your contribution!

[timminator]

Unfortunately the code style check/linter requires this ugly line break, so there is nothing I can do about it. :-/

[Bobholamovic]

Although the model directory name and the model name may be the same in some cases, they essentially refer to different things. Therefore, I don't think it's appropriate to use either the model directory name as the model name. Nevertheless, thank you for your contribution! If you have other ideas, feel free to share them with us.

[timminator]

But I checked the official models in PaddleX beforehand - when im not mistaken the model name and directory name matched in every instance.
Unless the user renames the folder, I cannot see how this could cause any issues. But even in that case it would result in the same error we are currently seeing in anyway, so it is not a regression and only makes it simpler most of the time.

[Bobholamovic]

But I checked the official models in PaddleX beforehand - when im not mistaken the model name and directory name matched in every instance. Unless the user renames the folder, I cannot see how this could cause any issues. But even in that case it would result in the same error we are currently seeing in anyway, so it is not a regression and only makes it simpler most of the time.

Yes, for official PaddleX models, the model name usually matches the directory name. However, users are free to rename the directory, so we can’t guarantee consistency there. To be more accurate, I recommend checking the model_name field in the inference.yml file inside the model directory. That’s the most reliable source. Feel free to contribute your code or open a feature request in the PaddleX GitHub repo!