This repository contains the official experiment implementation to the the TPAMI-2024 paper Surface Reconstruction Benchmark from point clouds: A Survey and a benchmark.
[Paper] [Dataset] [Project Page]
SurfaceRecBenchamrk
├──data # Put data here
| ├──synthetic_object
| ├──synthetic_scene
| └──real_object
├──build_dataset # Methods to build our benchmark datasets
| ├──scan_and_synthesis
| └──preprocessing
├──reconstruction # Reconstruction algorithms
|
└──metrics # Methods to evaluate the reconstructed surfaces
├──vanilla_metric
└──neural_metric
git clone https://github.com/Huang-ZhangJin/SurfaceRecBenchmark.git
git submodule update --init --recursive
There is a README.md file in each subfolder that describes how to use each script.
Download the Dataset and put it in the data folder
- To synthetic point clouds yourself:
- To perform object-level synthetic scanning, please follow instructions
- To perform scene-level synthetic scanning, please follow instructions
- Or use the point clouds provided by us
- To pre-processing the point clouds, please follow instructions
- Some surface reconstruction methods used in our paper
- Optionally, see our 3D Surface Reconstruction Library (Coming Soon!!!)
To use the following evaluation metrics, please follow instructions
- Vanilla metrics
- Chamfer Distance (CD)
- F-score
- Normal Consistency Score (NCS)
- Neural metrics
- Neural Feature Similarity (NFS)
If you find our work useful in your research, please consider citing:
@ARTICLE{
scutsurf_huang,
author={Huang, ZhangJin and Wen, Yuxin and Wang, ZiHao and Ren, Jinjuan and Jia, Kui},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
title={Surface Reconstruction from Point Clouds: A Survey and a Benchmark},
year={2024},
pages={1-20},
doi={10.1109/TPAMI.2024.3429209}
}