README.md

November 17, 2024 · View on GitHub

Real Sensors Dataset Description

This dataset folder contains a total of 1251 .pcd files, divided into 30 different categories. The data can be downloaded through [Tsinghua Cloud] or [Google Drive]. The data format is as follows:

├── bathtub/
│   ├──bathtub_000.pcd      
│   ├── bathtub_001.pcd       
│   ├── ...
│   └── bathtub_049.pcd
├── bed/
│   ├── bed_000.pcd
│   ├── bed_001.pcd
│   ├── ...
│   └── bed_049.pcd
├── ...
├── wardrobe/
│   └── wardrobe_000.pcd

Then put the downloaded folder to data/real_sensors_benchmark.

Application Method Description

Step 1: Point Cloud Completion

To run the test script, use the following command:

bash ./scripts/test.sh ${GPU_NAME} --ckpts ${CKPTS_PATH} --config ${CONFIG_PATH} --save_name ${SAVE_NAME}

Then the completed points will be saved and the Fidelity metric will be calculated.

Parameters:

  • GPU_NAME: specifies the name or number of the GPU to use.

  • CKPTS_PATH: specifies the path to the model checkpoint file that contains the model parameters saved at a specific training epoch.

  • CONFIG_PATH: specifies the path to the configuration file, which contains settings needed during training or testing.

  • SAVE_NAME: specifies the path name where the results will be saved.

Example

bash ./scripts/test.sh 0 --ckpts experiments/ProtoComp/PCN_models/example/pcn.pth --config cfgs/PCN_models/ProtoComp.yaml --save_path test

Step 2: Point Cloud Classification

The you can calculate Geometric Discriminability metric by running:

cd ./Pointnet_Pointnet2_pytorch
python test_classification.py --log_dir ${LOG_PATH} --data_path ${DATA_PATH}

Parameters:

  • ${LOG_PATH}: specifies the directory path which contains the (PointNet++) model checkpoint.

  • ${DATA_PATH}: specifies the directory path containing the results predicted by ProtoComp.

Example

python test_classification.py --log_dir pointnet2_cls_ssg --data_path ../data/real_sensors_benchmark

Additional Notes

  • Ensure all dependencies are installed, including libraries for point cloud processing and machine learning frameworks compatible with the pre-trained models.
  • Refer to the official documentation for detailed instructions on how to use the PointNet++ model.