Towards realistic symmetry-based completion of previously unseen point clouds

This is the anonymous demo code for the ICCV submission.

Installation

1. Clone the repo

   git clone https://github.com/ICCV2021Anonymous/symmetry_3D_completion.git

2. Install dependencies

   pip install -r requirements.txt

Usage

As the proof of concept we added the visualization of each step so it is recommended to run the following command:

  python completion.py --filename=models/shapenet/plane.pcd --visualization=y --time=y

Flags

• filepath - path to the damaged pcd file to complete
• visualization - (y/n) you can choose if you want the visualization of each step
• time - (y/n) print time report

Data

You can play with point clouds from different datasets:

• Damaged ShapeNet
• Damaged ModelNet
• Our new real-world dataset

Our full real-world dataset will be released with the camera-ready version.