Create README.md

README.md

@@ -0,0 +1,63 @@
+# Samba: Semantic Segmentation of Remotely Sensed Images with State Space Model
+
+This is the official code repository for "Samba: Semantic Segmentation of Remotely Sensed Images with State Space Model". {Arxiv Paper}
+
+![image](https://github.com/zhuqinfeng1999/Samba/assets/34743935/723109f3-4e5b-45c2-ad4f-492a87277075)
+
+## Abstract
+
+Remotely sensed images generally possess very high resolutions, making the semantic segmentation for remote sensing challenging. The commonly used methods for processing semantic segmentation are based on the Convolutional Neural Network (CNN) and Vision Transformer (ViT). However, due to the limitation of the receptive field in CNN, and the limitation of handling long sequences in ViT, it is difficult for the networks to handle high-resolution images. Inspired by Mamba, which is based on the State Space Model (SSM), capturing global semantic information with low computational complexity, we propose a semantic segmentation framework for remote sensing images, named Samba. The proposed framework utilizes an encoder-decoder architecture, employing Samba blocks as the encoder, which is designed for efficient image feature extraction to extract multi-level semantic information effectively, and UperNet as the decoder. We evaluate Samba on the LoveDA dataset, comparing it against top-performing CNN and ViT methods. The results reveal that Samba achieved unparalleled performance on LoveDA. This represents that the proposed Samba method is an effective application of the State Space Model in remote sensing image semantic segmentation, setting a benchmark for Mamba-based techniques in semantic segmentation of remotely sensed images. The source code and baseline implementations are available at https://github.com/zhuqinfeng1999/Samba.
+
+## Installation
+
+### Requirements
+
+Requirements: Ubuntu 20.04, CUDA 12.4
+
+* Set up the mmsegmentation environment; we conduct experiments using the mmsegmentation framework. Please refer to https://github.com/open-mmlab/mmsegmentation.
+* Install Mamba: execute pip install causal-conv1d==1.1.1 and pip install mamba-ssm
+* Install apex-amp, pyyaml, timm, tlt
+
+### Integrated into the mmseg framework
+
+#### LoveDA datasets
+
+* The LoveDA datasets can be found here https://github.com/Junjue-Wang/LoveDA.
+
+* After downloading the dataset, you are supposed to put them into '/mmsegmentation/data/loveDA/'
+
+* '/mmsegmentation/data/loveDA/'
+- ann_dir
+  - train
+    - .png
+  - val
+    - .png
+- img_dir
+  - train
+    - .png
+  - val
+    - .png
+
+#### Model file and config file
+
+- The model file Samba.py should be put into /mmsegmentation/
+
+- The config file samba_upernet.py for the combination of backbone and decoder head samba_upernet should be put into /mmsegmentation/configs/_base_/models
+
+- The config file samba_upernet-15k_loveda-512x512_6e4.py for training should be put into /mmsegmentation/configs/samba/
+
+## Training Samba
+
+`bash tools/dist_train.sh mmsegmentation/configs/samba/zqf_samba_upernet-15k_loveda-512x512_6e4.py 2 --work-dir mmsegmentation/output/sambaupernet --amp`
+
+## Testing Samba
+
+`bash tools/dist_test.sh /mmsegmentation/configs/samba/zqf_samba_upernet-15k_loveda-512x512_6e4.py \ /mmsegmentation/output/sambaupernet/iter_15000.pth 2 --out /mmsegmentation/visout/sambaupernet`
+
+## Paper
+
+
+
+## Acknowledgement
+
+We acknowledge all the authors of the employed public datasets, allowing the community to use these valuable resources for research purposes. We also thank the authors of Mamba, SiMBA, MMSegmentation for making their valuable code publicly available.