# DS-Vmamba

**Repository Path**: sst9/ds-vmamba

## Basic Information

- **Project Name**: DS-Vmamba
- **Description**: DS-Vmamba论文代码
- **Primary Language**: Unknown
- **License**: Apache-2.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2024-08-11
- **Last Updated**: 2025-04-28

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Ds-Vmamba：Dual-stream Network of Vision Mamba and CNN with Auto-scaling for Remote Sensing Image Segmentation :sunny: 

This repository will provide the code for the following paper:  :sunny: Dual-stream Network of Vision Mamba and
 CNN with Auto-scaling for Remote Sensing
 Image Segmentation:sunny:

> **Abstract:** 
# Code
Our code is based on MMSegmentaion (version 0.16.0).

## Installation
### Create a conda virtual environment and activate it (conda is optional)
```
conda create -n isdnet python=3.8 -y
conda activate isdnet
```
### Install dependencies
```
# Install pytorch firstly, the cudatoolkit version should be same in your system.
conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.1 -c pytorch

# Or you can install via pip
pip install torch==1.6.0 torchvision==0.7.0

# Install
python setup.py develop
```
## Datasets
We follow [FCtL](https://github.com/liqiokkk/FCtL) to split dataset.

### Inria Aerial
Please register and download [Inria Aerial](https://project.inria.fr/aerialimagelabeling/) dataset. 
Create folder named 'root_path', its structure is  
```
    root_path/
    ├── imgs
       ├── train
          ├── xxx_sat.tif
          ├── ...
       ├── test
       ├── val
    ├── labels
       ├── train
          ├── xxx_mask.png(two values:0-1)
          ├── ...
       ├── test
       ├── val
```

### Training
DeepGlobe
`./tools/dist_train.sh configs/isdnet/isdnet_r18-d8_1224x1224_80k_DeepGlobe.py 4`

Inria Aerial
`./tools/dist_train.sh configs/isdnet/isdnet_r18-d8_2500x2500_40k_InriaAerial.py 4`

### Evaluation
Accuracy:
`python tools/test.py config_file checkpoints_file --eval mIoU`

*Please download following pretrianed-model [here](https://drive.google.com/file/d/1FfG-qRlGy-2BsVjN2ZcKTEG9wZeo3sdW/view?usp=sharing)*

FPS:
`python tools/fps_test.py config_file --height height of the test image --width width of the test image`


## Results
### Potsdam
| Class |  building  | background |
| :---: | :---: | :---: |
| IoU | 74.39 | 97.58 |

### Inria Aerial
| Class |  building  | background |
| :---: | :---: | :---: |
| IoU | 74.39 | 97.58 |


## Citation
If you use this code and our results for your research, please cite our paper.
```