Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

UNetMamba: An Efficient UNet-Like Mamba for Semantic Segmentation of High-Resolution Remote Sensing Images

Enze Zhu, Zhan Chen, Dingkai Wang, Hanru Shi, Xiaoxuan Liu, Lei Wang

TL;DR

UNetMamba is a UNet-like semantic segmentation model based on Mamba that incorporates a Mamba segmentation decoder (MSD) that can efficiently decode the complex information within high-resolution images, and a local supervision module (LSM) which is train-only but can significantly enhance the perception of local contents.

Abstract

Semantic segmentation of high-resolution remote sensing images is vital in downstream applications such as land-cover mapping, urban planning and disaster assessment.Existing Transformer-based methods suffer from the constraint between accuracy and efficiency, while the recently proposed Mamba is renowned for being efficient. Therefore, to overcome the dilemma, we propose UNetMamba, a UNet-like semantic segmentation model based on Mamba. It incorporates a mamba segmentation decoder (MSD) that can efficiently decode the complex information within high-resolution images, and a local supervision module (LSM), which is train-only but can significantly enhance the perception of local contents. Extensive experiments demonstrate that UNetMamba outperforms the state-of-the-art methods with mIoU increased by 0.87% on LoveDA and 0.39% on ISPRS Vaihingen, while achieving high efficiency through the lightweight design, less memory footprint and reduced computational cost. The source code is available at https://github.com/EnzeZhu2001/UNetMamba.

UNetMamba: An Efficient UNet-Like Mamba for Semantic Segmentation of High-Resolution Remote Sensing Images

TL;DR

UNetMamba is a UNet-like semantic segmentation model based on Mamba that incorporates a Mamba segmentation decoder (MSD) that can efficiently decode the complex information within high-resolution images, and a local supervision module (LSM) which is train-only but can significantly enhance the perception of local contents.

Abstract

Semantic segmentation of high-resolution remote sensing images is vital in downstream applications such as land-cover mapping, urban planning and disaster assessment.Existing Transformer-based methods suffer from the constraint between accuracy and efficiency, while the recently proposed Mamba is renowned for being efficient. Therefore, to overcome the dilemma, we propose UNetMamba, a UNet-like semantic segmentation model based on Mamba. It incorporates a mamba segmentation decoder (MSD) that can efficiently decode the complex information within high-resolution images, and a local supervision module (LSM), which is train-only but can significantly enhance the perception of local contents. Extensive experiments demonstrate that UNetMamba outperforms the state-of-the-art methods with mIoU increased by 0.87% on LoveDA and 0.39% on ISPRS Vaihingen, while achieving high efficiency through the lightweight design, less memory footprint and reduced computational cost. The source code is available at https://github.com/EnzeZhu2001/UNetMamba.
Paper Structure (12 sections, 9 equations, 3 figures, 3 tables)

This paper contains 12 sections, 9 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. ResT Encoder
  4. Mamba Segmentation Decoder
  5. Local Supervision Module
  6. Loss Function
  7. Experimental results
  8. Datasets
  9. Experimental Settings
  10. Performance Comparison
  11. Ablation Studies
  12. Conclusion

Figures (3)

  • Figure 1: Framework of the proposed UNetMamba. (a) Overall architecture of UNetMamba. (b) Visual State Space (VSS) block of Mamba Segmentation Decoder (MSD). (c) Local Supervision Module (LSM) Block of LSM.
  • Figure 2: Qualitative comparison on the LoveDA dataset at resolution of 1024 × 1024 pixels. (a) Origin, (b) Ground Truth, (c) the proposed UNetMamba, (d) BANet, (e) MANet, (f) DC-Swin, (g) UNetFormer, (h) E-PyramidMamba, (i) CM-UNet and (j) RS$^3$Mamba.
  • Figure 3: Qualitative comparison on the ISPRS Vaihingen dataset at resolution of 1024 × 1024 pixels. (a) Origin, (b) Ground Truth, (c) the proposed UNetMamba, (d) BANet, (e) MANet, (f) DC-Swin, (g) UNetFormer, (h) E-PyramidMamba, (i) CM-UNet and (j) RS$^3$Mamba.