This is the code example for reproducing "Descriptor: Coastal Aerial Imagery Dataset for Shoreline Segmentation (CAID)".

CAID Dataset and Deep Learning Model Training

Download the CAID Dataset

Please download the CAID dataset from the following Dropbox link: Download CAID Dataset or Zenodo link: Download CAID Dataset.

Run DistributionAnalysis.ipynb to visualize dataset distribution.

Setting Up MMSegmentation

To test deep learning model performance on CAID, first download and install MMSegmentation.

Modify voc.py

After installation, modify the voc.py file located under mmseg/datasets to ensure it contains only two classes:

• background
• water

Also, update the color palette to:

PALETTE = [[0, 0, 0], [128, 0, 0]]

Organizing the Dataset

1. Navigate to the mmsegmentation folder.
2. Create a folder named datasets if it does not already exist.
3. Unzip the downloaded CAID dataset inside this datasets folder.
4. Rename the extracted folder to voc2012.

Copy my_model to mmsegmentation

Before training, copy the my_model folder to the mmsegmentation directory:

cp -r my_model mmsegmentation/

Then modify LINE 3 in pascal_voc12.py to your dataset path.

Training the Model

Running Training in Terminal

1. Open a terminal and activate the correct Python environment.
2. Run the following command to train CCNet on CAID:

   python tools/train.py --config ./my_model/CCNetBenchmark/ccnet_r50-d8_4xb4-20k_voc12aug-512x512.py --work-dir /home/weiwang/ResearchProjects/mmsegmentation/my_model_res/CCNetBenchmark/

3. To train other deep learning models provided in the my_model folder, run similar commands by replacing the --config and --work-dir arguments accordingly.
4. Once complete all the training, you can run DemoAnalysis.ipynb to see the performance on validation sets during training.

Performance Testing

Preparing the Test Set

1. Create a folder named test_set.
2. Copy all original images and labeled images into this test_set.
3. Inside test_set, create two subfolders:
   • img/ (for original images)
   • GT_label/ (for labeled images)

Running Model Inference

Follow the instructions in the MMSegmentation documentation to demo the trained model on the img folder inside test_set. Save the segmentation results into corresponding subfolders within test_set.

---

Once you have completed inferencing the test images and placed the results into the corresponding folders, you can follow BenchDemoAnalysis.ipynb and TestParamsAnalysis.ipynb to plot the examples and show the statistics.

By following these steps, you will successfully set up, train, and evaluate a deep learning model on the CAID dataset using MMSegmentation.

Images accessbility

The aerial imagery is provided by United States Department of Agriculture (USDA), which provides authorization for freely use. https://www.fsa.usda.gov/help/accessibility-statement

Contributors

Wei Wang, University of Wisconsin, Madison

Boyuan Lu, University of Wisconsin, Madison

Yihan Li, University of Wisconsin, Madison

Weiyan Shi, Singapore University of Technology and Design

Citation

@ARTICLE{11125715,
  author={Wang, Wei and Lu, Boyuan and Li, Yihan and Shi, Weiyan},
  journal={IEEE Data Descriptions}, 
  title={Descriptor: Coastal Aerial Imagery Dataset for Shoreline Segmentation (CAID)}, 
  year={2025},
  volume={},
  number={},
  pages={286 - 295},
  keywords={Sea measurements;Image segmentation;Benchmark testing;Accuracy;Labeling;Training;Lakes;Image edge detection;Boats;Algae;Aerial imagery;coastal;shoreline;segmentation;the Great Lakes},
  doi={10.1109/IEEEDATA.2025.3599116}}

W. Wang, B. Lu, Y. Li and W. Shi, "Descriptor: Coastal Aerial Imagery Dataset for Shoreline Segmentation (CAID)," in IEEE Data Descriptions, doi: 10.1109/IEEEDATA.2025.3599116.

Join our Kaggle Competition

We host a Kaggle competition at

Boyuan Lu and Wei Wang. Coastal shoreline segmentation. https://kaggle.com/competitions/coastal-shoreline-segmentation, 2025. Kaggle.