huoyijie / AdvancedEAST

AdvancedEAST is an algorithm used for Scene image text detect, which is primarily based on EAST, and the significant improvement was also made, which make long text predictions more accurate.https://github.com/huoyijie/raspberrypi-car
https://huoyijie.cn/
MIT License
1.23k stars 380 forks source link

quad invalid with vertex num less then 4 #102

Open keithyh opened 5 years ago

keithyh commented 5 years ago

用预训练模型会报这个警告信息,但是检测文本效果还可以(_act.jpg文件会找到所有文本区域,但是_predict.jpg会漏掉不少文本区域),请问这个警告信息是什么意思呢

misotrnka commented 4 years ago

I've noticed that this happens a lot when there are disjoint areas of text in the image.

For example, in this image, the algorithm cannot match the beginning of the second line with its end, because there are two disjoint regions. So some of the quadrangles are actually triangles, with two of the four points in [0, 0] (because this is how the array was initialized).

011 jpg_quads

As a workaround, you can alter the region_neighbor method in nms to be more forgiving (wider j_min, j_max and i_m).