DETSLideNet

Dual-branch UAV+Satellite multi-scale segmentation network

DETSlideNet fuses high-resolution UAV imagery and wide-area satellite data via cross-attention and a flexible decoder, for accurate segmentation across scales.

---

🚀 Features

• Dual-branch encoders (SwinTransformer) for UAV & SAT streams
• Cross-attention fusion at multiple scales
• Flexible decoder blocks with optional skip connections
• Joint loss combining UAV-supervision & SAT-supervision
• End-to-end PyTorch implementation with training & evaluation scripts

---

📁 Repository Structure

├── .github
│   └── workflows
│       └── ci.yml
├── checkpoints/           # saved model weights
├── data/                  # expected directory structure for datasets
│   ├── train/
│   └── test/
├── src/
│   ├── dataloader.py      # CASLDataset
│   ├── nets.py            # DETSLideNet and submodules
│   ├── train.py           # train_net
│   ├── test.py            # eval_net
│   └── utils/
│       ├── losses.py
│       └── module.py
├──  Dockerfile
├── .gitignore
├── .gitattributes
├── requirements.txt
└── README.md

⚙️ Installation & Usage

Follow the steps below to set up and run DETSLideNet, both locally (without Docker) and in Docker.

---

A. Prerequisites

• Python 3.8 or higher
• Git
• (Optional, for Docker) Docker & docker-compose

---

B. Clone the repository

git clone https://github.com/Guemann-ui/DETSlideNet.git
cd DETSlideNet

C. Data Download & Organization

Download the dataset from Zenodo.

Extract under data/:

data/
├── train/
│   ├── Moxitaidi/
│   │   ├── img/
│   │   └── label/
│   ├── TiburonPeninsula/
│   │   ├── img/
│   │   └── label/
│   └── Jiuzhaivalley/
│       ├── img/
│       └── label/
└── test/
    ├── Moxitaidi/
    │   ├── img/
    │   └── label/
    ├── TiburonPeninsula/
    │   ├── img/
    │   └── label/
    └── Jiuzhaivalley/
        ├── img/
        └── label/

D. 1. Local Setup (Without Docker)

1. Create & activate virtual env

python3 -m venv .venv
source .venv/bin/activate    # macOS/Linux
.venv\Scripts\activate       # Windows PowerShell

2. Install dependencies

pip install --upgrade pip
pip install -r requirements.txt

3. Train

python main.py \
  --mode train \
  --region Moxitaidi,TiburonPeninsula,Jiuzhaivalley \
  --epochs 300 \
  --batch-size 16 \
  --learning-rate 0.01 \
  --sat-weight 0.4 \
  --img_size 128 \
  --output-dir checkpoints/run1

4. Evaluate

python main.py \
  --mode test \
  --region Moxitaidi,TiburonPeninsula,Jiuzhaivalley \
  --load checkpoints/model_best.pth \
  --img_size 128

D. 2. Docker Setup

1. Build Docker image

docker build -t detslidenet:latest .

2. Train in container

PowerShell / Windows:

docker run --rm \
  --mount type=bind,source="$(pwd)/data",target=/app/data \
  --mount type=bind,source="$(pwd)/checkpoints",target=/app/checkpoints \
  detslidenet:latest \
  --mode train --epochs 100 --batch-size 4 --region Moxitaidi --img_size 128

Bash / Linux / macOS:

docker run --gpus all \
  -v $(pwd)/data:/app/data \
  -v $(pwd)/checkpoints:/app/checkpoints \
  detslidenet:latest \
  python main.py --mode train --output-dir checkpoints/run1

3. Evaluate in container

PowerShell / Windows:

docker run --rm `
  --mount type=bind,source="$($pwd.Path)\data\data",target=/app/data/data `
  detslidenet:latest `
  --mode test `
  --region Moxitaidi,TiburonPeninsula,Jiuzhaivalley `
  --load checkpoints/model_best.pth `
  --img_size 128

Bash / Linux / macOS:

docker run --gpus all \
  -v $(pwd)/data:/app/data \
  -v $(pwd)/checkpoints:/app/checkpoints \
  -v $(pwd)/results:/app/results \
  detslidenet:latest \
  python main.py --mode test --load checkpoints/run1/model_best.pth

4. (Optional) Docker Compose

docker-compose up --build

E. Quick Commands

python main.py -m train	Train with default settings
python main.py -m test -f <ckpt>	Evaluate saved checkpoint
docker run …	Train or test in Docker container