【论文阅读】Sliding Line Point Regression for Shape Robust Scene Text Detection

一、整体网络结构

        【论文阅读】Sliding Line Point Regression for Shape Robust Scene Text Detection     【论文阅读】Sliding Line Point Regression for Shape Robust Scene Text Detection

二、细节

     【论文阅读】Sliding Line Point Regression for Shape Robust Scene Text Detection            【论文阅读】Sliding Line Point Regression for Shape Robust Scene Text Detection          【论文阅读】Sliding Line Point Regression for Shape Robust Scene Text Detection          【论文阅读】Sliding Line Point Regression for Shape Robust Scene Text Detection

n=7,(7+7)*2+4=32个channel

三、结果

      【论文阅读】Sliding Line Point Regression for Shape Robust Scene Text Detection       【论文阅读】Sliding Line Point Regression for Shape Robust Scene Text Detection

四、总结

       1、不做连线的话只能水平方向

2、连线可以多方向,但可能是扭曲的不规则形状

3、两阶段的回归,提高准确率

五、问题

      1.回归的扭曲的多边形形状怎么算的f值???

上一篇:黄聪:C#如何Json转字符串;字符串转Json;Newtonsoft.Json(Json.Net)学习笔记(转)


下一篇:机器学习基石笔记:16 Three Learning Principles