Tensorflow implementation of PCT: Point Cloud Transformer.
Paper link: https://arxiv.org/pdf/2012.09688.pdf
The code is based on pointnet, pointnet2 and PCT_Pytorch
pointnet: https://github.com/charlesq34/pointnet
pointnet2: https://github.com/charlesq34/pointnet2
PCT_Pytorch: https://github.com/uyzhang/PCT_Pytorch
The data used in point cloud cls is ModelNet40 and split as pointnet.
After making tf_ops, downloading the modelnet40 and put it in datasets/modelnet40_ply_hdf5_2048.
Using train.py to train the PCT model, and testing via test.py.
The OA of my test results is 89%+, not as SOTA as paper demonstrated and other repo.
I doubt there exist some errors in my implementation.
If there exist any errors, please contact me via committing issue or liufeng@radi.ac.cn