README.md

---
license: cc-by-nc-4.0
task_categories:
- image-segmentation
- object-detection
---

# ❗ Latest Announcement: GTPBD Dataset & Paper (Updated on October 9th, 2025)

## 🎉 NeurIPS 2025 Acceptance
Our paper, **"GTPBD: A Fine-Grained Global Terraced Parcel and Boundary Dataset"**, has been officially accepted by **NeurIPS 2025**!  
The camera-ready version and the arXiv version will be released by **October 23, 2025**.  

## 📌 Before You Use GTPBD
If you plan to use the **GTPBD dataset**, please make sure to **carefully read the following notes and announcements**.  
They include important information about dataset versions, imagery sources, licensing compliance, and access instructions.

👉 This will help you correctly use the dataset, avoid common misunderstandings, and ensure compliance with data usage policies.

## ❗ Important Notice on Imagery and Google Earth TOS
To ensure full compliance with Google Earth’s Terms of Service:  
- **No Google Earth imagery** is distributed in this dataset.  
- For the small subset where GE was used, only **metadata (`.xml`) files** are provided.  
- These `.xml` files include geospatial info (coordinates, projection, extent), enabling users to reproduce the dataset legally.  

This approach guarantees **reproducibility** while respecting **data licensing policies**.  

## 📦 Dataset Update
- The dataset has been reconstructed for **higher quality and usability**.  
- The new release is **`GTPBD_enhenced_png`**, where most Google Earth imagery has been replaced with **GF-2 satellite imagery**.  
- Compared with the earlier `GTPBD_PNG` version, this update ensures improved consistency and reliability.  

### 🌍 Current Coverage
- **Replaced with GF-2 imagery**: majority of regions.  
- **Still using Google Earth (GE) imagery**: *South Africa, Brazil, Morocco, Nigeria, Tunisia*.  
  - For these five countries, we only provide **`.xml` metadata files** under the `Rest_of_world/` directory.  
  - Users can retrieve the corresponding imagery themselves via **Google Earth Pro** or other licensed high-resolution providers.  

## 📌 License and Usage

This dataset is released under the **Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)** license.

You are free to:

- Use and modify the annotation data for **non-commercial research and education**
- Share adapted versions of the data **with attribution**

You may **not** use the dataset or any part of it for commercial purposes. Please contact us if you are unsure whether your intended use qualifies.

---
# GTPBD: Global Terraced Parcel and Boundary Dataset (Updated Version)

Paper: [GTPBD: A Fine-Grained Global Terraced Parcel and Boundary Dataset](https://huggingface.co/papers/2507.14697)
Code: [https://github.com/Z-ZW-WXQ/GTPBD/](https://github.com/Z-ZW-WXQ/GTPBD/)

We are pleased to introduce **GTPBD**, the updated and enhanced version of our global terraced agricultural parcel dataset. This release supports fine-grained segmentation, edge detection, and geospatial learning tasks across diverse, often mountainous terrains where terrace farming is practiced.

**GTPBD** (Global Terraced Parcel and Boundary Dataset) is the first fine-grained benchmark dataset tailored for agricultural parcel extraction in **global terraced regions**. Unlike prior datasets focused on flat or regular farmlands, GTPBD specifically captures **complex terrain types**, providing **multi-level annotations** and supporting **multiple remote sensing tasks**, including:

- Semantic Segmentation  
- Edge Detection  
- Parcel Extraction  
- Unsupervised Domain Adaptation (UDA)  

# Statistics

*   **Geographic Coverage**: 7 major zones in China + global transcontinental regions
*   **Area Covered**: 885 km² of annotated terraces
*   **Label Diversity**:
    *   3-level labels (mask, boundary, parcel)
    *   Complex topologies (shared vs. non-shared field ridges)

## 📦 What's Included in `GTPBD`

The current release provides a refined subset of the dataset in `.png` format, named `GTPBD_enhenced_png`. It includes:

- ✅ Expanded annotation coverage, supporting more diverse terrain types and regions  
- ✅ Three-level manual labels: **parcel_labels**, **boundary_labels**, and **mask_labels**  
- ✅ `.xml` geospatial metadata files for each image, including:
  - Geographic coordinates (bounding box)
  - Spatial resolution
  - Projection information and image footprint  
These metadata files allow users to reference or re-acquire the corresponding satellite images for further geospatial analysis.

- **🌍 Global Scope**: 47,537 high-resolution samples from China, Vietnam, Tunisia, Ethiopia, Mexico, Peru, and more  
- **🌾 Terraced Focus**: 200,000+ annotated parcels with irregular, stepped, and shadowed structures  
- **🧩 Multi-Level Labels**:
  - Parcel Mask Labels  
  - Pixel-Level Boundary Labels  
  - Instance-Level Parcel Labels  
- **📑 Data augmentation for improved generalization in terraced parcel learning**:  
  - Images with **`rot`** in their filenames are **rotated versions** of the originals  
  - Images with **`flip`** in their filenames are **flipped versions** of the originals

# Tasks & Benchmarks

### 1. Semantic Segmentation

*   **Models**: U-Net, DeepLabV3, PSPNet, SegFormer, Mask2Former, etc.
*   **Metrics**: IoU, Pixel Accuracy, F1-score, Recall, Precision

### 2. Edge Detection

*   **Models**: UEAD, MuGE, PiDiNet, REAUNet-Sober
*   **Metrics**: ODS, OIS, AP

### 3. Agricultural Parcel Extraction

*   **Models**: HBGNet, SEANet, REAUNet
*   **Metrics**: OA, IoU, F1, plus object-level:
    *   GOC (Over-Classification Error)
    *   GUC (Under-Classification Error)
    *   GTC (Total Classification Error)

### 4. Unsupervised Domain Adaptation (UDA)

*   **Domains**: South (S), North (N), Global (G)
*   **Transfers**: S → N, G → S, etc.
*   **Methods**: Source Only, FDA, DAFormer, HRDA, PiPa

## Sample Usage

You can download the dataset directly from Hugging Face:

```bash
git lfs install
git clone https://huggingface.co/datasets/wxqzzw/GTD
```

## 📑 Explanation of Directories

The **GTPBD dataset** directory structure is illustrated below:

![GTPBD Directory Structure](dataset_structure.png)

- **Regional Subfolders**  
  - *Central China, East China, North China, Northeast China, Northwest China, South China, Southwest China*: Seven major Chinese agricultural regions.  
  - *Rest of the world*: Includes other international samples (e.g., South Africa, Brazil, Morocco, Nigeria, Tunisia).
---


## 📥 Need the Full `.tif` Satellite Imagery?

Most regions in **`GTPBD_enhenced_png`** already include the bundled **PNG** imagery.  
Only **five countries** currently do **not** include the imagery due to licensing constraints:
**South Africa, Brazil, Morocco, Nigeria, Tunisia**.

- For these countries, we provide **`.xml` metadata** under `Rest_of_world/` so you can legally retrieve the corresponding images from **Google Earth Pro** or other **licensed high-resolution providers**.
- If you **require original GeoTIFF (`.tif`)** for research use, please contact us directly.

For questions, feedback, or `.tif` access support, contact:  
📧 **zhangzhw65@mail2.sysu.edu.cn**