Converting-McGill-Dataset-to-ModelNet-Format

This repository provides code for transforming the McGill dataset into the ModelNet format to enable training on the McGill dataset using PointNet. (This repository provides code to create a McGill version of modelnet40_normal_resampled.)

PointNet: https://github.com/yanx27/Pointnet_Pointnet2_pytorch.git

Install Dataset

First, download the dataset(.ply) from the McGill 3D Shape Benchmark.

Second, extract each file(.gz) and organize the folders as follows.

project-root/
├── data/
│   ├── airplane/
│   │   ├── b1.ply
│   │   ├── b2.ply
│   │   ├── ...
│   ├── ant/
│   │   ├── 1.ply
│   │   ├── 2.ply
│   │   ├── ...
│   ├── ...
│   ├── teddy-bears/
│   │   ├── b1.ply
│   │   ├── b2.ply
│   │   ├── ...

Convert Dataset

You should download python library.

numpy, trimesh

pip install numpy
pip install trimesh

Edit base_path in McGill_convert.py. This is the folder for converted McGill dataset.

Run the McGill_convert.py.

python McGill_convert.py 

Finally, you can use the dataloader for modelnet40_normal_resampled directly with McGill.

(Train / Test split setting is "Local Spectral Graph Convolution for Point Set Feature Learning" in ECCV 2018.

project-root/
├── data/
│   ├── airplane/
│   │   ├── airplane_0000.txt
│   │   ├── airplane_0001.txt
│   │   ├── ...
│   ├── ant/
│   │   ├── ant_0000.txt
│   │   ├── ant_0001.txt
│   │   ├── ...
│   ├── ...
│   ├── teddy-bears/
│   │   ├── teddy-bears_0000.txt
│   │   ├── teddy-bears_0001.txt
│   │   ├── ...
|   ├── mcgill_shape_names.txt
|   ├── train.txt
|   ├── test.txt