Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

请问一下层次hard mining体现在代码的哪里? #1

Open
iridescent303 opened this issue Nov 1, 2017 · 9 comments
Open

请问一下层次hard mining体现在代码的哪里? #1

iridescent303 opened this issue Nov 1, 2017 · 9 comments

Comments

@iridescent303
Copy link

之前看技术圈的解释,没看懂。

@YiYuanIntelligent
Copy link
Owner

hardmining 的代码在这块。https://github.com/YiYuanIntelligent/3DFasterRCNN_LungNoduleDetector/blob/master/lib/rpn/tianchi_anchor_target_layer.py#L97 。层次化hard mining是工程上的一种实现,没有放在网络里面,我们会先训练一个模型,作为第一层模型。然后以这个模型的基础上,finetune比上一层模型更难分的数据,然后作为第二层模型,依次往下。

@iridescent303
Copy link
Author

哦,是把第一次得到的bbox放到第二次进行训练是吧

@iridescent303
Copy link
Author

对了,你们这里的batchnorm层后面为什么不加scale层啊,感觉这样和pytorch里面的不一样了。

@YiYuanIntelligent
Copy link
Owner

我记得当时,逐层核对过caffe和pytorch的输出,batchnorm这块的实现,两者有挺大的差异。我们应该是尝试过,加或不加,caffe都无法正常收敛,所以后面没管这层了

@iridescent303
Copy link
Author

你们固定了输入比较了输出吗,我实现的加了scale输出是一样的。

@YiYuanIntelligent
Copy link
Owner

150层的结果都一样吗?bn那块的参数是怎么设置的

@iridescent303
Copy link
Author

都用的默认的参数,不过我是保持全部权重一致。

@iridescent303
Copy link
Author

不过bn层参数caffe和pytorch的不太一样

@zp678
Copy link

zp678 commented Apr 24, 2018

Hello, I want to know how to create a data set for this network. I didn't participate in the Tianchi competition so I can't download it. Can you share it?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants