Simple-Remote-Sensing-Change-Detection-Framework

Introduction

This project is a simplified implementation of remote sensing change detection based on pytorch, I hope it can help those who are beginners in change detection domain implementing their ideas quickly, without concerning other things. It has the following features:

Using albumentations to implement abundant data augmentations.
Using wandb to log hyper-parameters, metrics and images, so that we can analyse experiment and adjust hyper-parameter easily.
Using torchmetrics to compute metrics quickly and properly.
Using warn up, amp, and other basic training strategies.
Many dataset process functions such as crop image, random split image are provided, as some change detection dataset needs to be processed with our own needs.
Abundant annotations are provided in most of functions and classes.
Code and its logic are very simple, easy to understand.
Change detection model is a simple twin network, which served as an example, and loss function is just addition of BCELoss and DiceLoss.

Install dependencies

Return the following code in command line.

pip install -r requirements.txt

Data

Using any change detection dataset you want, but organize dataset path as follows. dataset_name is name of change detection dataset, you can set whatever you want.

dataset_name
├─train
│  ├─label
│  ├─t1
│  └─t2
├─val
│  ├─label
│  ├─t1
│  └─t2
└─test
    ├─label
    ├─t1
    └─t2

Below are some binary change detection dataset you may want.

WHU Building

Paper: Fully convolutional networks for multisource building extraction from an open aerial and satellite imagery data set

DSIFN

Paper: A deeply supervised image fusion network for change detection in high resolution bi-temporal remote sensing images

LEVIR-CD

Paper: A Spatial-Temporal Attention-Based Method and a New Dataset for Remote Sensing Image Change Detection

LEVIR-CD+

GoogleMap

Paper: SemiCDNet: A Semisupervised Convolutional Neural Network for Change Detection in High Resolution Remote-Sensing Images

SYSU-CD

Paper: SYSU-CD: A new change detection dataset in "A Deeply-supervised Attention Metric-based Network and an Open Aerial Image Dataset for Remote Sensing Change Detection"

CDD

Paper: CHANGE DETECTION IN REMOTE SENSING IMAGES USING CONDITIONAL ADVERSARIAL NETWORKS

NJDS

Paper: Semantic feature-constrained multitask siamese network for building change detection in high-spatial-resolution remote sensing imagery

S2Looking

Paper: S2Looking: A Satellite Side-Looking Dataset for Building Change Detection

Start

For training, run the following code in command line.

python train.py

If you want to debug while training, run the following code in command line.

python -m ipdb train.py

For test and inference, run the following code in command line.

python inference.py

简介

此项目是一个遥感变化检测的极简代码框架，希望能够帮助到刚进入变化检测领域的人快速实现自己的想法，代码主要包含以下特点：

使用albumentations库进行丰富的数据增强
使用wandb记录下每次实验的超参数、训练和验证指标、训练和验证结果图片，保证每次实验过后都可以详细的分析原因，并且也不用额外在其他地方记录超参数，调参的时候也更方便
使用torchmetrics库来快速并且正确地计算指标，避免了初期我单独计算每个batch的指标再求平均的错误
使用了warm-up、amp等基本的训练策略
某些变化检测数据集需要自己进行额外的数据处理，比如对一整幅遥感影像进行裁剪、把遥感图像随机分配到训练集、验证集和测试集等，写好了各种数据集处理的代码，基本足够各种情况下的使用了
用尽可能规范的方式写了详尽的注释，基本代码逻辑很容易看懂
代码量少，代码逻辑清晰，尽可能用简洁的代码进行了实现
变化检测模型是一个简单的孪生网络结构，损失函数就是简单的bce损失和dice损失的和

下载需要的库

下载CUDA
下载1.12或者更新的pytorch
下载其他需要的包

在命令行中运行下面的命令下载其他需要的包