search

unmannedlab / LiDARNet

LiDARNet: A Boundary-Aware Domain Adaptation Model for Point Cloud Semantic Segmentation
https://unmannedlab.github.io/research/LiDARNet
32 stars 3 forks source link

code #2

Closed kongxin123456 closed 3 years ago

kongxin123456 commented 3 years ago

Hi, Peng Jiang. Could you please share the code with me? I want to see some specific details of the code and learn from it. Thank you!

maskjp commented 3 years ago

Hi, @kongxin123456

The code is not ready to release. I made some changes and my coding style is not good. If you really want to read the code, I can send you a version which might be difficult to read.

Sorry for that.

Best wishes!

kongxin123456 commented 3 years ago

Hi, @kongxin123456

The code is not ready to release. I made some changes and my coding style is not good. If you really want to read the code, I can send you a version which might be difficult to read.

Sorry for that.

Best wishes!

Thank you for your quick reply. I would like to learn the specific settings of the network through the code. In addition, I also want to learn about the processing of multiple data sets. Please send me a copy if it is convenient for you. My email is kongxin@nuaa.edu.cn, or you can provide me with a download link.

kongxin123456 commented 3 years ago

I have one more question. Does the SemanticUSL still have only the sequences 3, 12, 21, and 32 labeled while other sequences are not labeled? If I use SemanticUSL as the source domain data set for training, can I only use the sequence 3,12,21,32? I see that each sequence in the data set has a label folder.

maskjp commented 3 years ago

I have one more question. Does the SemanticUSL still have only the sequences 3, 12, 21, and 32 labeled while other sequences are not labeled? If I use SemanticUSL as the source domain data set for training, can I only use the sequence 3,12,21,32? I see that each sequence in the data set has a label folder.

Yes, only 3,12,21,32 have ground truth. You can use this tool (https://github.com/jbehley/point_labeler) to visualize the data or label your new sequence. You can train a model on the labeled sequences and make predictions for other sequences. And then you can correct the labels using the above tools.

There are some other datasets available for you to evaluate your own method now, such as

http://www.cvlibs.net/datasets/kitti-360/

https://www.nuscenes.org/nuscenes

By the way, I send a version of code to you through email. Please check it.

Hope these information helps!

Best wishes!

kongxin123456 commented 3 years ago

Sincerely thank you for your help. I have received your email and code, and I will only use the code for personal study and will not share it with others.