The image annotation before json.dumps() encoding is a list containing multiple dictionaries. The `points` in the dictionary represent the coordinates (x, y) of the four points of the text box, arranged clockwise from the point at the upper left corner.
The image annotation after json.dumps() encoding is a list containing multiple dictionaries. The `points` in the dictionary represent the coordinates (x, y) of the four points of the text box, arranged clockwise from the point at the upper left corner.
`transcription` represents the text of the current text box, and this information is not needed in the text detection task.
If you want to train PaddleOCR on other datasets, you can build the annotation file according to the above format.