An Empirical Study of Remote Sensing Pretraining

Di Wang, Jing Zhang, Bo Du, Gui-Song Xia and Dacheng Tao

Updates | Introduction | Usage | Results & Models | Talk | Statement |

Current applications

Scene Recognition: Please see Remote Sensing Pretraining for Scene Recognition;

Sementic Segmentation: Please see Remote Sensing Pretraining for Semantic Segmentation;

Object Detection: Please see Remote Sensing Pretraining for Object Detection;

Change Detection: Please see Remote Sensing Pretraining for Change Detection;

ViTAE: Please see ViTAE-Transformer;

Matting: Please see ViTAE-Transformer for matting;

Updates

018/10/2023

RSP won the highly cited paper!

027/05/2022

The early access is available!

020/05/2022

The paper has been accepted by IEEE TGRS.

011/04/2022

The baiduyun links of pretrained models are provided.

07/04/2022

The paper is post on arxiv!

06/04/2022

The pretrained models for ResNet-50, Swin-T and ViTAEv2-S are released. The code for pretraining and downstream tasks are also provided for reference.

Introduction

This repository contains codes, models and test results for the paper "An Empirical Study of Remote Sensing Pretraining".

The aerial images are usually obtained by a camera in a birdview perspective lying on the planes or satellites, perceiving a large scope of land uses and land covers, whose scene is usually difficult to be interpreted since the interference of the scene-irrelevant regions and the complicated spatial distribution of land objects. Although deep learning has largely reshaped remote sensing research for aerial image understanding and made a great success. However, most of existing deep models are initialized with ImageNet pretrained weights, where the natural images inevitably presents a large domain gap relative to the aerial images, probably limiting the finetuning performance on downstream aerial scene tasks. This issue motivates us to conduct an empirical study of remote sensing pretraining. To this end, we train different networks from scratch with the help of the largest remote sensing scene recognition dataset up to now-MillionAID, to obtain the remote sensing pretrained backbones, including both convolutional neural networks (CNN) and vision transformers such as Swin and ViTAE, which have shown promising performance on computer vision tasks. Then, we investigate the impact of ImageNet pretraining (IMP) and RSP on a series of downstream tasks including scene recognition, semantic segmentation, object detection, and change detection using the CNN and vision transformers backbones.

Fig. - (a) and (b) are the natural image and aerial image belonging to the "park" category. (c) and (d) are two aerial images from the "school" category. Despite the distinct view difference of (a) and (b), (b) contains the playground that is unusual in the park scenes but usually exists in the school scenes like (d). On the other hand, (c) and (d) show different colors as well as significantly different spatial distributions of land objects like playground and swimming pool.

Results and Models

MillionAID

Backbone	Input size	Acc@1	Acc@5	Param(M)	Pretrained model
RSP-ResNet-50-E300	224 × 224	98.99	99.82	23.6	google & baidu
RSP-Swin-T-E300	224 × 224	98.59	99.88	27.6	google & baidu
RSP-ViTAEv2-S-E100	224 × 224	98.97	99.88	18.8	google & baidu

Usage

Please refer to Readme.md for installation, dataset preparation, training and inference.

Citation

If this repo is useful for your research, please consider citation

@ARTICLE{rsp,
  author={Wang, Di and Zhang, Jing and Du, Bo and Xia, Gui-Song and Tao, Dacheng},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={An Empirical Study of Remote Sensing Pretraining}, 
  year={2023},
  volume={61},
  number={},
  pages={1-20},
  doi={10.1109/TGRS.2022.3176603}
}

Talk

A video talk about this study (In Chinese)

Statement

This project is under MIT licence. For any other questions please contact di.wang at gmail.com or di_wang at whu.edu.cn.

Relevant Projects

[1] Advancing Plain Vision Transformer Towards Remote Sensing Foundation Model, IEEE TGRS, 2022 | Paper | Github
     Di Wang^∗, Qiming Zhang^∗, Yufei Xu^∗, Jing Zhang, Bo Du, Dacheng Tao and Liangpei Zhang