Whats the reason behind switching to the server models in 3.0.1?

[timminator]

In the 3.0.0 release the default one was PP-OCRv5_mobile and in 2.10 it was for example PP-OCRv4 mobile even though at that time there were also server models available.
The server level models are considerably larger and also considerably slower on CPU, so I'm thinking about switching back to the PP-OCRv5 mobile model in my own project and I was just curious what the reasons were for making this relatively big change.

Thanks for taking the time to read my question.

[cuicheng01]

This is a very interesting question, and there are several main reasons behind it:

• We’ve noticed that many users are now using GPUs for inference, which allows us to set the default model to the server version.

• Many new users who are not familiar with PaddleOCR may find that the mobile model doesn’t meet their expectations the first time they use it. Since they might not dig deep into the documentation, they may not realize that there’s also a server model available, and therefore miss out on experiencing PaddleOCR’s more powerful OCR capabilities.

• Our online demo uses the server model by default, so when many users try the default mobile model locally and see a significant difference compared to the online experience, it can lead to confusion and questions.

[timminator]

Thanks! That answers my question.