/O-CNN

O-CNN: Octree-based Convolutional Neural Networks for 3D Shape Analysis

Primary LanguageC++MIT LicenseMIT

O-CNN

This repository contains the implementation of our papers related with O-CNN.
The code is released under the MIT license.

If you use our code or models, please cite our paper.

Contents

What's New?

  • 2021.08.24: Update the code for pythorch-based O-CNN, including a UNet and some other major components. Our vanilla implementation without any tricks on ScanNet dataset achieves 76.2 mIoU on the ScanNet benchmark, even surpassing the recent state-of-art approaches published in CVPR 2021 and ICCV 2021.
  • 2021.03.01: Update the code for pytorch-based O-CNN, including a ResNet and some important modules.
  • 2021.02.08: Release the code for ShapeNet segmentation with HRNet.
  • 2021.02.03: Release the code for ModelNet40 classification with HRNet.
  • 2020.10.12: Release the initial version of our O-CNN under PyTorch. The code has been tested with the classification task.
  • 2020.08.16: We released our code for 3D unsupervised learning. We provided a unified network architecture for generic shape analysis tasks and an unsupervised method to pretrain the network. Our method achieved state-of-the-art performance on several benchmarks.
  • 2020.08.12: We released our code for Partnet segmentation. We achieved an average IoU of 58.4, significantly better than PointNet (IoU: 35.6), PointNet++ (IoU: 42.5), SpiderCNN (IoU: 37.0), and PointCNN(IoU: 46.5).
  • 2020.08.05: We released our code for shape completion. We proposed a simple yet efficient network and output-guided skip connections for 3D completion, which achieved state-of-the-art performances on several benchmarks.

Please contact us (Peng-Shuai Wang wangps@hotmail.com, Yang Liu yangliu@microsoft.com ) if you have any problems about our implementation.