https://github.com/wzh99/DCP-TF

SJTU CS473 Project: Implementation of Deep Closest Point in TensorFlow, and its comparison with other registration methods.
https://github.com/wzh99/DCP-TF

deep-learning point-cloud-registration tensorflow tensorflow-graphics

Last synced: 4 days ago
JSON representation

SJTU CS473 Project: Implementation of Deep Closest Point in TensorFlow, and its comparison with other registration methods.

• Host: GitHub
• URL: https://github.com/wzh99/DCP-TF
• Owner: wzh99
• Created: 2020-06-01T10:37:57.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2020-06-14T08:20:40.000Z (over 4 years ago)
• Last Synced: 2024-08-01T19:46:49.977Z (3 months ago)
• Topics: deep-learning, point-cloud-registration, tensorflow, tensorflow-graphics
• Language: C++
• Homepage:
• Size: 8.34 MB
• Stars: 12
• Watchers: 3
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome-cs - @wzh99, 2020 Spring

README

        
# Deep Closest Point in TensorFlow

## Introduction

This project implements [Deep Closest Point](https://arxiv.org/abs/1905.03304) model in TensorFlow. It also includes C++ code that compare its performance with other registration methods (ICP, 4-PCS, Go-ICP).

## Dependencies

To run DCP model, you may have to install these Python packages:

* tensorflow>=2.0.0

* tensorflow-graphics (none of its dependencies is required)

* numpy

* h5py

To run comparison program, you may have to install these libraries:

* PCL 1.9 (and its dependencies)

* HDF5

* TBB

## Usage

Basic usage is encapsulated into procedures. You can directly call them in the program. Hyperparameters are directly defined in source code, and command line arguments is not supported.

### Dataset

Download [ModelNet40](https://shapenet.cs.stanford.edu/media/modelnet40_ply_hdf5_2048.zip) and unzip files into directory `modelnet40`. Run `util.pack_to_one()` to pack all dataset files into single `train.h5` and `test.h5` files. 

### Training and evaluation

Trained weights `dcp_v2.h5` can be unzipped from [`weights/dcp_v2.zip`](weights/dcp_v2.zip). Place it in `weights` directory so that evaluation and testing procedure can find it. If you want to train by yourself, run `train.train()` to train, or your owning training procedure. Run `train.evaluate()` to evaluate the trained model with test dataset.

### Comparison

The comparison program tests registration methods on the first 100 models of the test dataset. It is divided into Python and C++ code. Run `compare.test_dcp()` to test DCP. Compile and run the C++ program to test ICP, 4-PCS and Go-ICP. ICP and 4-PCS implementation is from PCL. Go-ICP is from my previous project [OptICP](https://github.com/wzh99/OptICP).

## Documentation

The project [proposal](doc/proposal.md) and [report](doc/dcp_report.md) are provided (both in Chinese). Refer to them for better understanding of this project.