VI-PointConv-pytorch

This repository contains the official implementation of The Devils in the Point Clouds: Studying the Robustness of Point Cloud Convolutions in PyTorch.

It is tested under CUDA 10.2, and PyTorch 1.6. The hardware is a single V100 GPU for the ScanNet experiment.

A. Environment Setup

1. Use Anaconda

conda env create -f environment.yml

B. Prepare Data

1, Please refer https://github.com/facebookresearch/SparseConvNet/blob/master/examples/ScanNet/prepare_data.py to process the raw data

2, Open './dataGeneration/scannet_data_generateNorms.py', please specify the paths for the raw data ({train_files, val_files, test_files}) from the previous step, and the corresponding saving paths {save_train_path, save_val_path, save_test_path}. Then run

python scannet_data_generateNorms.py

C. Training & Evaluation

1. Training

All hyperparameters are set within ./config.yaml, please read through and change them correspondingly

python train_ScanNet.py

2. Evaluation

python test_ScanNet.py