[Feature] how to get text-detection contours，not bounding box？

[a819411321]

Motivation

I think mmdeploy_point_t should be contours,not four points.
In many cases, we do not directly use text recognition after text detection.
The text recognition model has good recognition performance for non curved text, but poor recognition performance for curved text. Therefore, when performing text recognition on a curved section of text, it is best to first remove the curvature. However, the current text detection results only return the coordinates of four points in the text area. In many cases, it is not possible to correct curved text based on these four points, so contour information of the text is required.

Related resources

No response

Additional context

No response