/classification3dmodel

3D model classification

Primary LanguageJupyter Notebook

Classification3dmodel

This is a Jupyter notebook for classifying 3DCAD models using CNN. I am using Keras as a framework.

Requirement

Keras  2.2.4
tensorboard 1.9.0     
tensorflow  1.9.0     
tensorflow-gpu  1.10.0
numpy 1.14.5
matplotlib  3.0.0 
scikit-learn  0.20.0
tqdm  4.26.0

Corresponding file format

  • OFF
  • STL

How to use

  • 3DCAD models data download
    ModelNet10/40 downlod

    $ git clone https://github.com/tacky0612/classification3dmodel.git
$ cd classification3dmodel
$ bash get_modelnet.sh

  • Convert 3DCAD model to voxels
    Please run "make_npy_file.ipynb" with Jupyter notebook.
    It converts the 3D data to voxels. Then, t is saved as numpy

  • Voxel visualization
    Please run "vis.ipynb" with Jupyter notebook. It makes visualization of 3D model.
    ↓ Like this ↓

  • Classification 3Dmodels Please run "Classification_3DSAMPLE.ipynb" with Jupyter notebook.
    It classifies 3D models.

    Using this architecture.

    ↓ Accuracy ↓

    ModelNet10 ModelNet40 My dataset(3D_SAMPLE)
    90.1% 86.8% 97.6%

When this does not work

Please check ./config.py
'DATASET = Modelnet10 or Modelnet40'
Remove the comment out of this．
3D_SAMPLE is our own dataset. I can not publish it now.

P.S.

Forgive me for my poor English... XD

My twitter ← Please contact me if you have any troubles.