SAIL: The Spatial AI Library

SAIL (Spatial AI Library) is an open-source framework for building, training, and explaining spatial deep learning models — designed to make GeoAI accessible to both experts and newcomers.

It provides a modular configuration-based workflow for:

• Training models on satellite imagery and geospatial data
• Evaluating and validating spatial predictions
• Running explainability and perturbation analyses (e.g., SIMBA, SIA)
• Integrating new models, samplers, and datasets with minimal code

---

Quick Start

Install directly from GitHub:

pip install git+https://github.com/heatherbaier/sail.git

Run your first experiment:

python launch.py --config configs/phl_geoconv_regression.yaml

This example trains a GeoConv-based regression model on imagery and tabular data defined in the configuration file.

All outputs — models, logs, and metrics — will be saved to the specified output_dir in your config.

Configuration-Based Design

Every experiment in SAIL is defined through a YAML configuration file that specifies:

• Dataset source and parameters
• Model architecture and initialization
• Training hyperparameters
• Validation and explainability settings

Example:

task: train
experiment_name: "usaid_dhs_pct_water_improved"
output_dir: "artifacts/checkpoints/phl_geoconv_income"

dataset:
  type: json
  data_root: "/data/phl"
  prefix: "pct_improved_water_africa"
  batch_size: 16
  img_size: [224, 224]

model:
  name: geoconv
  params:
    pretrained: false

trainer:
  epochs: 1
  lr: 1e-4
  device: "cuda"

Documentation

Full usage instructions, tutorials, and developer guides are available in the SAIL Wiki: https://github.com/heatherbaier/sail/wiki

There, you'll find:

• How to structure and run experiments
• How to register new models or datasets
• How to extend SAIL with explainability or spatial robustness modules
• Examples and notebooks for GeoAI research and teaching