Evaluation Metrics of Dual Octree Graph Networks

This repository contains the evaluation metrics of our paper Dual Octree Graph Networks, which are implemented by Convolutional Occupancy Networks.

Installation

• Create an anaconda environment called pytorch-1.4.0 using

  conda env create -f environment.yaml
  conda activate pytorch-1.4.0
  

• Compile the extension modules.

  python setup.py build_ext --inplace
  

Evaluation

Denote the folder where you clone the code of our dual octree graph networks as ognn.

• Evaluate the results on the testing dataset of ShapeNet.

  python eval_meshes.py  \
        configs/pointcloud/shapenet.yaml  \
        --dataset_folder /ognn/data/ShapeNet/dataset  \
        --generation_dir /ognn/logs/shapenet_eval/test
  

• Evaluate the results on the unseen 5 categories of ShapeNet.

  python eval_meshes.py  \
        configs/pointcloud/shapenet.yaml  \
        --dataset_folder /ognn/data/ShapeNet/dataset.unseen5  \
        --generation_dir /ognn/logs/shapenet_eval/unseen5
  

• Evaluate the results on the synthetic room dataset.

  python eval_meshes.py  \
        configs/pointcloud/room.yaml  \
        --dataset_folder /ognn/data/room/synthetic_room_dataset  \
        --generation_dir /ognn/logs/docnn/room_eval/room