Skip to content

Kyle1993/simplest_FasterRcnn_pytorch

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

28 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Simplest implement of FasterRCNN by Pytorch

Base on chenyuntc/simple-faster-rcnn-pytorch

1、重新组织了代码结构,方便阅读
2、添加了中文注释和自己的理解
3、 修改了部分代码

  • rpn的score输出只有一维概率,loss改成BCE
  • rnp_loc_loss 只计算正样本计算
  • extractor&rpn 和 roi_head分开训练,先训练若干epoch的extractor&rpn
  • 删除smooth_l1_loss的sigma

4、make it fit Pytorch 0.4+
5、用pytorch自带的AdaptiveMaxPool2d替换原来ROIPool2D,代码更简洁易懂,但是慢!!!(https://www.cnblogs.com/king-lps/p/9026798.html)
6、删除了cupy计算MNS的部分,纯numpy实现,代码更简洁易懂,但是 慢!!!
7、默认只支持batch_size = 1,把不必要的部分删去
8、fix some bugs

Tips:
Adam效果比SDG差

TODO:
still have some bugs, can't predict correctly
支持多batch_size
添加Focal Loss

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages