Classification

ModelNet40

  1. Clone the ocnn-pytorch repository, and enter the subdirectory projects. Run the following command to prepare the dataset.

    python tools/cls_modelnet.py

  2. Train the LeNet used in our paper O-CNN. The classification accuracy on the testing set without voting is 91.7%. And the training log and weights can be downloaded here.

    python classification.py --config configs/cls_m40.yaml SOLVER.alias time

  3. Train the HRNet used in our paper on 3D Unsupervised Pretraining. The classification accuracy on the testing set without voting is 93.0%. And the training log and weights can be downloaded here.

    python classification.py --config configs/cls_m40_hrnet.yaml SOLVER.alias time