EllipsoidQuery

This code submission is to reproduce the impact of re-oriented ellipsoid querying on RS-CNN Shape Classification.

contact email: ssheshap@udel.edu

Software requirements

Ubuntu 18.04
Python 3.5 (recommend Anaconda3)
Pytorch 0.4.1
CMake 3.10.2
CUDA 10.0 + cuDNN 7
Cudatoolkit V10.0.130

Note: Also, works in the environment suggested by the authors of RS-CNN(https://github.com/Yochengliu/Relation-Shape-CNN/).

Download

git clone https://github.com/VimsLab/EllipsoidQuery.git
cd EllipsoidQuery/RSCNNEQ

Building Kernel

mkdir build && cd build
cmake .. && make

Dataset

Download and unzip ModelNet40 (415M) in data directory.
https://shapenet.cs.stanford.edu/media/modelnet40_ply_hdf5_2048.zip

Usage: Train

sh train_cls.sh

Note: We have trained a Single-Scale-Neighborhood classification model in cls folder, whose training accuracy is 92.55% ('cls/model_cls_ssn_iter_70917_acc_0.925446.pth')

Usage: Evaluation

Modify cfgs/config_ssn_cls.yaml with *.pth file from cls/ folder with highest accuracy.
python voting_evaluate_cls.py

Note: You can use our model cls/model_cls_ssn_iter_70917_acc_0.925446.pth as the checkpoint in config_ssn_cls.yaml, and with majority voting you will get an accuracy of 93.51%. Due to randomness the accuracy might vary.

This code has been heaviy borrowed from https://github.com/Yochengliu/Relation-Shape-CNN/ and https://github.com/erikwijmans/Pointnet2_PyTorch

To cite our paper please use below bibtex.

        @InProceedings{Sheshappanavar_2020_CVPR_Workshops,
            author = {Venkanna Sheshappanavar, Shivanand and Kambhamettu, Chandra},
            title = {A Novel Local Geometry Capture in PointNet++ for 3D Classification},
            booktitle = {The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
            month = {June},
            year = {2020}
        }