c++ - 对截断文本的 opencv EAST 检测的高置信度得分

Question

我正在使用 opencv EAST 检测的这个 c++ 示例源代码 ( text_detection.cpp ) 来检测这个截止文本

这是结果。我还在每个检测到的文本框下方绘制了准确度。

所有检测到的文本框的置信度都很高，如何调整程序使其返回较低的置信度或对上图没有检测结果？

标签： c++opencv

Answer 1

Looking at the image, I'd call it a true positive. There is definitely text in the boxes drawn. Well, digits, but that's not the sort of difference you get from EAST.

You might be confusing EAST with an OCR algorithm; it isn't.