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.
The `points` in the dictionary represent the multi-point coordinates (such as: 4 points, 8 points and 14 points, etc.) of the text box, arranged clockwise from the point at the upper left corner.
`transcription` represents the text of the current text box. **When its content is "###" it means that the text box is invalid and will be skipped during training.**