The source code of PRA-Net.
- CUDA >= 10
- python = 3.7
- pytorch = 1.2.0
- numpy = 1.16.4
- Classification
- Download ModelNet40, and unzip it to
data/modelnet40_ply_hdf5_2048 - Download dataset from ScanObjectNN Projcet, and unzip it to
data/h5_files
- Download ModelNet40, and unzip it to
-
Before running the training script, you need to compile some customized torch operators.
cd _ext-src python setup.py install
-
To train a PR-Net model on the ScanObjectNN dataset:
CUDA_VISIBLE_DEVICES=0 python train_cls.py --config cfg/ScanObjectNN_train_PRANet.yaml
Our code is mainly based on DGCNN, ScanobjectNN and RSCNN, thanks for them!