oliveirabruno01/blind-cartographer-data

Blind Cartographer Dataset

This repository contains the ground-truth data for the "Blind Cartographer" experiment, inspired by the blog post "How Does A Blind Model See The Earth?" by Henry.

The dataset provides a high-resolution global grid of coordinates (0.5° per point), with each point labeled as either land (1) or water (0).

This dataset is intended for use in evaluating or training language models on geographical knowledge and spatial reasoning.

Notebook used to generate this dataset: Open in Colab

Data Source & Acknowledgements

The ground-truth data for this dataset was derived from the Natural Earth project, a public domain map dataset available at naturalearthdata.com.

Specifically, we used the 10m Gray Earth (GRAY_HR_SR_OB_DR) raster dataset.

Made with Natural Earth.

Dataset Structure

The dataset has a single configuration and a single split: train.

This split contains the full, high-resolution grid of 259,200 points. Lower resolutions for training or evaluation should be created by downsampling this grid on the fly. For robust results, this is best achieved via aggregation: grouping all the high-resolution points that fall within a larger grid cell and using a majority rule to determine if that cell is land or water.

How to Use

You can easily load the dataset using the datasets library:

from datasets import load_dataset

# Load the high-resolution grid
dataset = load_dataset("oliveirabruno01/blind-cartographer-data", split="train")

# Access the first data point
print(dataset[0])
# Expected output:
# {'longitude': -179.75, 'latitude': 89.75, 'is_land': 0}

Data Fields

Each data point has the following simple structure:

• longitude (float): The longitude of the coordinate, ranging from -180.0 to 180.0.
• latitude (float): The latitude of the coordinate, ranging from -90.0 to 90.0.
• is_land (int): A binary flag indicating the surface type.
  • 1: The coordinate is over land, ice, or snow.
  • 0: The coordinate is over water.

Citation

If you use this dataset, please consider giving credit to the original experiment by https://x.com/arithmoquine:

@misc{henry2025blindmodel,
  author       = {Henry},
  title        = {How Does A Blind Model See The Earth?},
  year         = {2025},
  url          = {https://outsidetext.substack.com/p/how-does-a-blind-model-see-the-earth},
  note         = {Accessed: 2025-08-30}
}