Official implementation of the paper HoliTracer: Holistic Vectorization of Geographic Objects from Large-Size Remote Sensing Imagery.
Abstract: This paper introduces HoliTracer, the framework designed to holistically extract vectorized geographic objects from large-size RSI. In HoliTracer, we enhance segmentation of large-size RSI using the Context Attention Net (CAN), which employs a local-to-global attention mechanism to capture contextual dependencies. Furthermore, we achieve holistic vectorization through a robust pipeline that leverages the Mask Contour Reformer (MCR) to reconstruct polygons and the Polygon Sequence Tracer (PST) to trace vertices. Extensive experiments on large-size RSI datasets, including buildings, water bodies, and roads, demonstrate that HoliTracer outperforms state-of-the-art methods.
- OS: Linux distribution, our paper experiments are based on Ubuntu 22.04
- Hardware: At least one GPU with 24GB memory and CUDA support, our paper experiments are based on NVIDIA A100 GPUs 40GB
git clone https://github.com/vvangfaye/HoliTracer.git
cd HoliTracer
pip/conda install torch torchvision # our paper experiments are based on pytorch 2.5.1
pip install -r requirements.txt # install other dependencies
# install pycocotools with holitracer compatible version.
git clone https://github.com/vvangfaye/cocoapi-holi.git
cd cocoapi-holi/PythonAPI && python setup.py install
# install holitracer
cd ../../ && pip install -e . # install holitracer with editable mode| Dataset Name | Image Size | Spatial Resolution | Images | Train/Val/Test | Download Link |
|---|---|---|---|---|---|
| WHU-building | 10,000 × 10,000 | 0.075 m | 400 | 320 / 40 / 40 | Google Drive |
| GLH-water | 12,800 × 12,800 | 0.3 m | 250 | 200 / 25 / 25 | Google Drive |
| VHR-road | 12,500 × 12,500 | 0.2 m | 208 | 166 / 21 / 21 | Google Drive |
Download the datasets from the provided links and extract them to the data/datasets directory.
Pre-trained models and performance metrics:
| Dataset | PoLiS ↓ | CIoU | AP | APs | APm | APl | IoU | F1 | Download Link |
|---|---|---|---|---|---|---|---|---|---|
| WHU-building | 3.63 | 82.30 | 61.07 | 40.37 | 80.30 | 60.00 | 91.60 | 95.41 | Google Drive |
| GLH-water | 81.87 | 59.24 | 20.84 | 19.88 | 38.77 | 72.29 | 85.68 | 91.51 | Google Drive |
| VHR-road | 134.13 | 6.10 | 1.58 | 0.08 | 0.40 | 3.99 | 46.48 | 60.63 | Google Drive |
Download the pre-trained models from the provided links and extract them to the data/models directory.
Run the demo file to get started:
- demo.py and geo_demo.py: Includes examples for inference and visualization.
Training scripts and instructions are available in:
Refer to this file for detailed steps to train HoliTracer on target dataset.
If you have any questions about it, please let me know. (Create an 🐛 issue or 📧 email: wangfaye@whu.edu.cn)
We are developing a unified vectorization framework for remote sensing imagery in EarthVec, and we are happy to collaborate with you.