Mango-Mars/MF-Mamba

# MF-Mamba: Multiscale Convolution and Mamba Fusion Model for Semantic Segmentation of Remote Sensing Imagery This is an official implementation of MF-Mamba in our TGRS 2025 paper "MF-Mamba: Multiscale Convolution and Mamba Fusion Model for Semantic Segmentation of Remote Sensing Imagery". ## Main Environments ``` conda create -n MFMamba python=3.8 conda activate MFMamba pip install torch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 --index-url download.pytorch.org pip install causal_conv1d==1.0.0 pip install mamba_ssm==1.0.1 ``` If problems occur during installation, the corresponding precompiled package can be downloaded and used for installation. Refer to causal-conv1d and mamba-ssm. ## Pretrained weights We use the ImageNet pretrained HRNet-W18-C model 'hrnetv2_w18_imagenet_pretrained.pth' from HRNets. ## Citation If you find it useful, please consider citing: ``` @ARTICLE{11098811, author={Xiao, Pu and Dong, Yuting and Zhao, Ji and Peng, Tieqi and Geiß, Christian and Zhong, Yanfei and Taubenböck, Hannes}, journal={IEEE Transactions on Geoscience and Remote Sensing}, title={MF-Mamba: Multiscale Convolution and Mamba Fusion Model for Semantic Segmentation of Remote Sensing Imagery}, year={2025}, } ```