A Computation-aware Shape Loss Function for Point Cloud Completion

This repository is still under development, and we have updated a demo version for everyone to verify its functionality.

The complete codebase will be updated soon. If you have any questions, please feel free to contact us via email.

Data set for this small demo

10 different point clouds, repeat 8 times for each.

Run AAIP compare with emd2.py to see our improvement.

Average Time Cost:

AAIP: 0.7046 emd2: 0.1686

Average MSE(small better):

AAIP_MSE: 373.468 emd2_MSE: 3120.06

Usage

1) Envrionment & prerequisites

Pytorch 1.12.1
CUDA 11.6
Python 3.10
Visdom
Open3D

2) Compile

Compile our EMD modules:

    cd emd
    python3 setup.py install

3) Compare AAIP and emd2

    python AAIP compare with emd2.py

4) Output log file

emd_value_comparison.txt
time_cost_comparison.txt

5) Calculate MSE

    g++ -O3 -o comparison_MSE1 cal_MSE.cpp
    ./comparison_MSE1

6) License

This project Code is released under the Apache License 2.0 (refer to the LICENSE file for details).

7) Acknowledgments

Our code is partially based on MSN-Point-Cloud-Completion by Liu et al. We greatly appreciate their outstanding contribution. You can explore their code through the following link to appreciate their work.

coldbubbletea/AAIP-Point-Cloud-Completion