PointCNN

CNN Generalized for Consuming Point Cloud Data

View on GitHub

Datasets Preparation

Download

If you want to download :

You can use download_datasets.py:

python download_datasets.py -f [path to data folder] -d [Dataset to download]

For Scannet, please refer to http://www.scan-net.org/ . In segmentation task, We follow pointnet++ preprocessed data (Onedrive link).

For S3DIS, please refer to http://buildingparser.stanford.edu/dataset.html#Download

Converting to .h5 Files

For big scene point cloud datasets like Scannet and S3DIS, we split them into small blocks for training:

python3 split_data/scannet_split.py
python3 split_data/s3dis_prepare_label.py
python3 split_data/s3dis_split.py

Then, the .h5 files can be generated by:

python3 prepare_[dataset]_data.py -f [Path to data folder]

If you want to use extra features, such as RGB, you can use:

python3 prepare_multiChannel_seg_data.py -f [Path to data folder] -c [Channel number]