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

双任务学习有关问题 #15

Open
yimishuang opened this issue Jan 16, 2017 · 6 comments
Open

双任务学习有关问题 #15

yimishuang opened this issue Jan 16, 2017 · 6 comments

Comments

@yimishuang
Copy link

你好!非常感谢你的程序,让我在人脸特征点检测上面得到很多启发。关于多任务学习现有如下疑问:
1.看到你有两个cov4,之后分别处理了68个特征点的线性回归,和3Dpose线性回归,关于标签在最后损失层具体是怎么处理的?因为没有看到你加slice层,所以有些困惑~
2.能否告知一下你数据内容划分?标签那里是68个点的坐标+3个方向值么?

再次感谢你的研究,对我的帮助非常大。若有时间还望告知解答谢谢!

@qiexing
Copy link
Owner

qiexing commented Jan 19, 2017

@yimishuang
在处理数据用,我是将输入存储成hdf5的格式的,所以没有slice。
标签是68个点的坐标+3个方向值。

@yimishuang
Copy link
Author

感谢你在百忙之中赐教~
可是我还是想知道你的输入数据里面是包含了三项内容么?就是(Crop_Image)+(68piont_lable)+(3Dpose)?还是说你做了两个HDF5文件,分别是(Crop_Image)+(68piont_lable),(Crop_Image)+(3Dpose)?
能不能麻烦你把数据转换成HDF5格式的程序也共享一下么?我从网上找到了一个5个特征点检测的hdf5转换程序,但是对于多任务训练,我不知道该怎么处理了~

最后,马上就要过年了,提前拜个年!哈哈!

@qiexing
Copy link
Owner

qiexing commented Jan 23, 2017

@yimishuang
对的,一个文件包含3个信息:
hdf5也与prototxt对应
top: "data" 对应crop image
top: "label" 对应68point
top: "pose" 对应3dpose
你可以看看这个:https://github.com/qiexing/caffe-regression/blob/master/kaggle_fkp_data/kfkd.py
里面有怎么将数据存成hdf5的

@yimishuang
Copy link
Author

好的,非常感谢!

@Yimishuangshuang
Copy link

你好,我看了您关于训练数据集方面问题的解答,能问一下你训练30000+数据大约需要多久么?最大迭代次数按照1700000来算的话?

@qiexing
Copy link
Owner

qiexing commented Feb 8, 2017

@Yimishuangshuang 具体记不清了啊,大概得要3,4天?

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