From SAM to Mamba: A Remote Sensing Segmentation Archive

What it does

This repository gathers the PyTorch implementations of a research group’s semantic segmentation work for remote sensing. It targets very-high-resolution aerial and satellite imagery across several distinct technical tracks: multimodal fusion, unsupervised domain adaptation, SAM-based segmentation, and Mamba-based state-space models.

The interesting bit

Instead of evolving a single codebase, the repo operates as an anthology of published IEEE papers. Several included works carry ESI “Highly Cited” or “Hot Paper” badges, indicating the implementations map to methods that have gained real traction in the remote sensing literature.

Key highlights

• Covers diverse architectures: SAM-assisted segmentation (SAM_RS, MFNet), Mamba state-space models (RS^3Mamba), and GAN-based unsupervised domain adaptation (GLGAN, MBATA_GAN)
• Targets standard remote sensing datasets such as ISPRS Potsdam, Vaihingen, and LoveDA
• Includes utility scripts for image splitting, merging, and loss-curve plotting
• Multiple included papers are ESI Highly Cited or Hot Papers

Caveats

• The README is essentially a bibliography; it offers little guidance on how the methods relate or which to pick for a specific task
• The repo appears to be a collection of separate paper implementations rather than a unified framework with a consistent API
• One contact email link is malformed, suggesting maintenance attention goes to paper releases over polish

Verdict

Remote sensing researchers looking to reproduce or benchmark recent segmentation papers will find this a convenient consolidated starting point. If you need a single, actively maintained library with a stable interface, this is not it.