/City-Facade

Dataset and Code for City-Facade

City-Facade-logo

City-Facade

This is the Dataset and Code of City-Facade proposed by our paper "City-Facade: Segmentation of Building Facade Elements in Large-Scale 3D Point Clouds".

City-Facade is a large-scale urban building facade dataset for semantic-level and instance-level segmentation from MLS LiDAR point clouds. It consists of labeled point clouds (with 9 classes for building facades) as well as unlabeled data (point clouds of street landscapes). The data collection area encompasses a variety of streets in Xiamen, China, with distinct architectural styles. We believe that our City-Facade will faciliate feature research on point cloud semantic or instance segmentation, urban understanding and modeling, point cloud completion, etc.

City-Facade Preview

Xiamen

CAR

JMC

XHR

YWR

Dataset Download

Please enjoy this dataset we have provided. Click here to download small examples.

Installation

Requirements

  • Python 3.6.0 or above
  • Pytorch 1.2.0 or above
  • CUDA 10.0 or above

Virtual Environment

source install.sh

Data Preparation

(1) View and Download the City-Facade training / testing for building facade segmentation.

(2) Put the data in the corresponding folders, which are organized as follows.

data
|--City-Facade
|  |--train
|  |  |--CAR
|  |  |--JMC
|  |  |--SGS
|  |  |--XHR
|  |  |--YWR
|  |--test
|  |  |--CAR
|  |  |--JMC
|  |  |--SGS
|  |  |--XHR
|  |  |--YWR
...

Model Zoo

Method Model OA mIoU
PointNet PointNet 74.57 11.87
PointNet++ PointNet++ 74.76 11.82
DGCNN DGCNN 75.77 11.70
DeepGCNs DeepGCNs 76.04 11.95
ASSANet ASSANet-L 81.61 34.49
PointTrans. PointTrans. 74.83 39.26
PointNeXt PointNeXt-L 78.79 22.19
PointVector PointVector 85.22 36.57
PointMetaBase PointMetaBase-L 86.27 39.34

Organizations

SYSUXMUWater