This article focuses on the powerful 3D point cloud classification function of PointNet++, and explains in detail how to train your own data set. Before that, you need to ensure that you can run the training and testing of the source code. If not, please refer to the source code of PointNet ++ .
data placement
1.1. Configure your own classification and classification name in mytensor_shape_names.txt:
1.2. Fill in the corresponding folder name/data file name of different categories in filelist.txt, the folder name here should be consistent with your own classification category
1.3. Then enter the names of the training and testing data files in sequence in mytensor_train.txt and mytensor_test.txt
My training set is the test set, so it's the same
1.4 Actual test data collected again
code modification
2.1 Code modification during training
Mainly modify the batch_size, and the number of categories
In the data loading module, put your own configuration file in
Then you can start training
2.2 Code modification during testing
There are many modifications during the test, mainly in the data loading class, you can directly view the code
Run result view
To check whether your classification is correct, you can directly view the printed pred_choice
4. Source code download
You can go to Baidu Netdisk to download and test, if you have any questions, you can private message me
Link: https://pan.baidu.com/s/1iuqHQg_w6kq6JzmZjaIbaw
Extraction code: ng3h