Skip to content

ChenJiangxi/Urban-Green-Community-Segmentation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

High-Resolution Network Based Urban Green Community Space Segmentation

Introduction

HRNet (High-Resolution Network) has achieved high scores on many tasks of semantic segmentation and instance segmentation. And HRNetV2 was proposed for semantic segmentation. This work applied auxiliary learning on the basis of HRNetV2, using two auxiliary decoders added to the main decoder. The structure of aux-HRNet is illustrated as follows.

structure

Dataset

We produced a dichotomous remote sensing dataset named 'singlegreen' for the identification of urban community green spaces (UCGS).

An example of a labeled image is shown below.


rawdata

labeldata
In labeldata, UCGS is marked in red(RGB=[128,0,0]), and black(RGB=[0,0,0]) is the background. And the structure of our dataset is modeled after the cityscapes dataset. The split.py file is used to obtain the filenames of the images in the training dataset and stored separately in the splits folder.
├── singlegreen
│   ├── rawdata
│   │   ├── train
│   │   ├── val
|   |   ├── test
│   ├── labeldata
│   │   ├── train
│   │   ├── val
|   |   ├── test
|   ├── splits
│   │   ├── train.txt
│   │   ├── val.txt
|   |   ├── test.txt
├── split.py

Installation

Our work is based on the outstanding framework of MMsegmentation, and many thanks to all contributors. Please refer to get_started.md for installation.

Get Started

Please see train.md and inference.md for the basic usage of MMSegmentation. There are also tutorials for customizing dataset, designing data pipeline, customizing modules, and customizing runtime. We also provide many training tricks for better training and useful tools for deployment.

When dataset is prepared, you can train aux-HRNet using:

python tools/train.py configs/hrnet/fcn_aux-hr48_256x512_80k_singlegreen.py

inference the trained model:

python tools/test.py configs/hrnet/fcn_aux-hr48_256x512_80k_singlegreen.py work_dirs/fcn_hr48_256x512_80k_123+23aux_[0.4,0.4]/xx.pth --eval mIoU

Link to the trained model on singlegreen dataset: Trained Model

Method Backbone Precision Recall IoU F1score OA
FCN aux-HRNet-W48 83.01 85.69 72.91 84.33 89.31
FCN HRNet-W48 82.05 85.76 72.21 83.87 88.92
FCN aux-HRNet-W18 84.46 83.3 72.23 83.88 89.24
FCN HRNet-W18 78.58 88.92 71.57 83.43 88.14

Citation

If you find this project useful in your research, please consider cite:

waiting

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages