Experiments of the paper Variance-Aware Weight Initialization for Point Convolutional Neural Networks

This is the code of the experiments presented in the paper: Variance-Aware Weight Initialization for Point Convolutional Neural Networks.

NOTE: This repository only contains the code of the experiments. The CUDA implementation of the different convolution operators and the point cloud library in Pytorch needs to be installed separately from the following repository. See the Required libraries section in this file.

Required libraries

This code was tested in python 3.7. In order to install the required python libraries use pip with the following command:

    pip install -r requirements.txt

PCCNN Library

The library to process point clouds in Pytorch (including the convolution operators) can be found in the following repository. Install this library first in order to run the experiments.

Data Download

ScanObjectNN

This data set can be downloaded here. Once downloaded, uncompress the file in the ScanObjectNN folder and rename the folder to data.

ScanNet

This data set can be downloaded here.

ModelNet40

The preprocessed dataset can be downloaded from the following link. Once downloaded, uncompress the file in the ModelNet40 folder.

Run Experiments

In order to run the experiments, first you will need to create the folder where the log files will be stored. This is the code to create this folder for the ScanObjectNN task. For the other experiments we will need to create the same folder.

    cd ModelNet40
    mkdir runs

Once this folder is created, we can execute the bash script to run the different configurations (w/o our weight init scheme, w/o batch norm, w/o group norm, etc.).

    sh train_script.sh

If we want to use a different convolution operation, we need to modify the bash script and change the input parameter --conv_type with a valid convolution operation: mcconv, kpconv, kpconvn, pccnn, pointconv, or sphconv.

Once the training is finished we can validate our models with the following command substituting PATH by the path to the save model.

    python val.py --model_path PATH

Citation

If you find this code useful please consider citing us:

    @article{hermosilla2022weightinit,
      title={Variance-Aware Weight Initialization for Point Convolutional Neural Networks},
      author={Hermosilla, Pedro and Schelling, Michael and Ritschel, Tobias and Ropinski, Timo},
      journal={European Conference on Computer Vision (ECCV)},
      year={2022}
    }