haomo-ai / MotionSeg3D

[IROS 2022] Efficient Spatial-Temporal Information Fusion for LiDAR-Based 3D Moving Object Segmentation
https://npucvr.github.io/MotionSeg3D/
GNU General Public License v3.0
242 stars 21 forks source link

how to get the moving labels? #15

Closed beyounged closed 1 year ago

beyounged commented 1 year ago

hi, I have a problem about how to get the labels only moving (not unlabeled or static)?

MaxChanger commented 1 year ago

Sorry, can you explain the problem further? Do you mean how to label new data? Or how to visualize data labels?

beyounged commented 1 year ago

作者你好,我的意思是你的labels应该是一个三分类的labels(unlabeled or static or moving)当使用.raw.yaml的时候,因为输出shape是[bs,3,64,2048] ?labels应该是给点云打上[0,1,2]的标签(即proj_labels的取值是0,1,2中之一?)那能否单独拿只有moving的labels?

MaxChanger commented 1 year ago
  1. So proj_labels is one of 0,1,2, right?
    Yes, This is the same as the mapping to the right of learning_map in yaml.
  2. Can you take the moving labels separately? Do you mean slicing the output labels? such as pred_label[..., 1:2, :, :].

If I understand wrong, please describe in more detail

beyounged commented 1 year ago

Hi,抱歉回复晚了,关于第二点,我想问是否可以单独将proj_labels里的动态标签[2]保留,然后其余标签(静态/unlabeled)都为[0]?换句话说,是否提供了只有动态类别的标签?如果没有,应该如何制作?

MaxChanger commented 1 year ago

Hello, what you are talking about should be the slicing operation in python. It should be OK to mask the predicted label according to the value.

beyounged commented 1 year ago

ok, thanks for your quick reply, I am interested in this project, it's a wonderful work!

MaxChanger commented 1 year ago

Thank you, I will close this issue, if you have other questions, welcome to open a new issue.