Word-wise detection handwritten

[PxCPeter2801]

I have a dataset handwritten had annotated, and I used PPOCRLabel.py to test detect capabilities. I met 2 problems:

• How to rate the efficiency of the model when PPOCRLabel detects a line ( ex: 1 line includes 10 words, it is detected as 2 bounding boxes instead of 1 bounding box). So I think detection word-wise (1 word 1 box) is more efficient than line-wise
• How I change the model is used in PPOCRLabel.py, i want to apply my own model trained with handwritten

[tink2123]

1. The default detection model is line-level. If you want to use word-level detection, you can change the English detection model.
2. For instructions on changing models, please refer to the documentation: https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.7/PPOCRLabel/README.md#32-built-in-model

[PxCPeter2801]

I met a problem when auto recognition images handwritten English. Image and bounding box look like rotation 90 degrees. This case met some images, not at all

[SWHL]

This is due to an error in recognizing the direction of the current text line