---
license: apache-2.0
tags:
- salient-object-detection
- remote-sensing
- computer-vision
- pytorch
---

<a id="top"></a>
<div align="center">
  <h1>🚀 AESINet: Adaptive Edge-aware Semantic Interaction Network for Salient Object Detection in Optical Remote Sensing Images</h1>

  <p>
    <b>Xiangyu Zeng</b><sup>1</sup>&nbsp;
    <b>Mingzhu Xu</b><sup>1</sup>&nbsp;
    <b>Yijun Hu</b><sup>1</sup>&nbsp;
    <b>Haoyu Tang</b><sup>1</sup>&nbsp;
    <b>Yupeng Hu</b><sup>1✉</sup>&nbsp;
    <b>Liqiang Nie</b><sup>1</sup>
  </p>

  <p>
    <sup>1</sup>Affiliation (Please update if needed)
  </p>
</div>

Official implementation of **AESINet**, an Adaptive Edge-aware Semantic Interaction Network for Salient Object Detection (SOD) in Optical Remote Sensing Images.

🔗 **Journal:** IEEE Transactions on Geoscience and Remote Sensing (TGRS), 2023  
🔗 **Task:** Salient Object Detection (SOD)  
🔗 **Framework:** PyTorch  

---

## 📌 Model Information

### 1. Model Name
**AESINet** (Adaptive Edge-aware Semantic Interaction Network)

---

### 2. Task Type & Applicable Tasks
- **Task Type:** Salient Object Detection / Remote Sensing  
- **Core Task:** Salient object detection in optical remote sensing imagery  
- **Applicable Scenarios:**
  - Remote sensing scene understanding  
  - Aerial image analysis  
  - Environmental monitoring  

---

### 3. Project Introduction

Salient Object Detection (SOD) in optical remote sensing images is challenging due to complex backgrounds, low contrast, and ambiguous object boundaries.

**AESINet** introduces an **adaptive edge-aware semantic interaction mechanism**, which:

- Enhances edge-aware feature representation  
- Promotes multi-level semantic interaction  
- Improves boundary localization accuracy  
- Strengthens robustness under complex backgrounds  

---

### 4. Training Data

Common datasets used in remote sensing SOD:

- ORSSD  
- EORSSD  
- ORSI  

---

## 🚀 Pre-trained Weights

### Model Weights

- **AESINet-V (VGG backbone):**  
  https://pan.baidu.com/s/1Xo97lQF4TS2jak9v8iU8jA?pwd=qegm  

- **AESINet-R (ResNet backbone):**  
  https://pan.baidu.com/s/1gYW9qOjR0YjU5R4dCN9Hfg?pwd=tj25  

---

### Backbone Pretrained Weights

- **VGG & ResNet:**  
  https://pan.baidu.com/s/18k9e3YcxK1rTY8A_WajTyg?pwd=lb8l  

---

## 🚀 Training & Testing

### Step 1: Prepare Data
- Download datasets and pre-trained weights  
- Place them into corresponding directories  

### Step 2: Generate Dataset Lists
```bash
python generateTrainList.py
python generateTestList.py

### Step 3: Configure Paths
Modify dataset paths in the code
Ensure .txt file paths are correctly set

## ⚠️ Notes
Both ResNet and VGG versions are available
Code readability will be improved in future updates
GPU is recommended for training and inference

## 📝 Citation
@ARTICLE{10198281,
  author={Zeng, Xiangyu and Xu, Mingzhu and Hu, Yijun and Tang, Haoyu and Hu, Yupeng and Nie, Liqiang},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={Adaptive Edge-Aware Semantic Interaction Network for Salient Object Detection in Optical Remote Sensing Images}, 
  year={2023},
  volume={61},
  pages={1-16},
  doi={10.1109/TGRS.2023.3300317}
}

## 📬 Contact

For any questions, please contact:

## 📧 z15264367990@163.com