/O-CNN

This repository contains the code of our O-CNN paper.

Primary LanguageC++

Fork of O-CNN: Octree-based Convolutional Neural Networks

Virtual Scanner

cd O-CNN/virtual_scanner/.

This folder contains the code for converting 3D models to dense point clouds with normals. It outputs a mesh in the PLY format.

Install

apt-get install -y --no-install-recommends libboost-all-dev libcgal-dev libeigen3-dev
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make

Usage

    VirtualScanner <file_name> [nviews] [flags] [normalize]
        file_name: the name of the file (*.obj; *.off) to be processed.
        nviews: the number of views for scanning. Default: 6
        flags: Indicate whether to output normal flipping flag. Default: 0
        normalize: Indicate whether to normalize input mesh. Default: 0
Example:
    VirtualScanner input.obj 30         // process the file input.obj

Easily convert ply to obj

apt-get install assimp-utils
assimp export test.ply test.obj

Point shuffling

Shuffle points sampled from Virtual Scanner. Can be useful to quickly load a subsampled version on the pointcloud (just read a random block of lines in a PLY).

Dependencies : pip install argparse joblib and install Pymesh


#Usage on a folder of files
python randomizePointCloud.py --shapenet_path path

Both in one line

python process_raw_obj.py --shapenet_path path