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

坐标归一化 #30

Open
Crispus1119 opened this issue Dec 28, 2019 · 5 comments
Open

坐标归一化 #30

Crispus1119 opened this issue Dec 28, 2019 · 5 comments

Comments

@Crispus1119
Copy link

您好,我想问一下坐标归一化是什么意思

@yizt
Copy link
Owner

yizt commented Dec 29, 2019 via email

@Crispus1119
Copy link
Author

谢谢您的回答,试图读您的源码,但是对
# 坐标归一化 rois /= tf.constant(self.image_max_dim, dtype=tf.float32)
这句不解,为什么要对回归值除以?看了遍论文都么有对应的解释。很不解,如果可以的话希望您能详细的解释下这句代码,十分感谢!

@yizt
Copy link
Owner

yizt commented Dec 29, 2019

@Crispus1119您好, tf.image.crop_and_resize的boxes参数接受的是归一化后的边框坐标。

@Crispus1119
Copy link
Author

感谢回复,
# RPN和RCNN网络边框回归标准差 RPN_BBOX_STD_DEV = np.array([0.1, 0.1, 0.2, 0.2]) BBOX_STD_DEV = np.array([0.1, 0.1, 0.2, 0.2])
还有最后一个问题想问您,这些是如何得到的和标准化会起到什么样的作用。谢谢!!!

@yizt
Copy link
Owner

yizt commented Dec 29, 2019

@Crispus1119 标准差是一个经验值;标准化有两个作用,一个加速训练,二是让网络更易训练;如以下两个情况;
a) 回归的目标值都很小如:-0.1~0.1直接;那么梯度必然也小;训练自然慢
b) 回归的目标的两个值一个很小如~0.1~0.1,一个很大如-1000~1000;这样的值范围差距很大的情况;一般机器学习都要做标准化;所谓标准化就是减去均值再除以标准差。

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

2 participants