王权富贵：RCNN的成长历程

• 暴力目标检测                                                                                                 

  •  滑动窗口（从左到右，从上到下）（多个尺寸大小）                               

•   输入图片归一化                                                                                                                      

 

 

 

• R-CNN                                                                                                            

  •  选择性搜索                                                                                            

•  ROI图片归一化                                                                                     
•  使用选择搜索有风险，合并太大，会损失目标（损失的是重叠目标）。合并太小，数据太多                                                                     

总：

 

扫描二维码关注公众号，回复： 4279174 查看本文章

 

 

 

 

• Fast R-CNN                                                                                                     

•  边界框回归器                                                                                         

•  在卷积特征提取后使用Selectiv Search搜索方法                                     

•  特征图块归一化（ROI池化）                                                                

 

 

•  Faster R-CNN                                                                 

•  候选区域网络（RPN）                                                                           

 

 

 

 

• YOLO                                       

  •  训练回归框                      
  •  暴力设S*S的网络           
    •  太大不精确               
    •  太小分类不好做        

 

 

 

•  SSD                                                          

•  多尺度检测                             

•  卷积测物，加框先验