Library for AI-Assisted Mapping Tool developed for Humanitarian OpenStreetMap Team

A small team from Omdena worked on a disaster management project. This package was created in order to simplify the integration of the data processing steps with the model training one.

data Directory Structure

.
├───images
│   ├───1
│   ├───2
│   ├───3
│   ├───4
│   └───5
├───inputs
│   ├───1
│   ├───2
│   ├───3
│   ├───4
│   └───5
├───masks
│   ├───1
│   ├───2
│   ├───3
│   ├───4
│   └───5
└───predictions
    ├───1
    ├───2
    ├───3
    ├───4
    └───5

• inputs: GeoJSON labels and image files given to us.
• images: Georeferenced images with the fourth band removed (if any).
• masks: Rasterized labels.
• predictions: Masks predicted by some ML model.

API Reference

1. preprocess(data_path, input_dir, image_dir, mask_dir)

   Function for fully preprocessing the input data.

   • data_path: Path of the directory where all the data are stored.
   • input_dir: Name of the directory where the input data are stored.
   • image_dir: Name of the directory where the images are stored.
   • mask_dir: Name of the directory where the masks are stored.

2. predict(checkpoint_path, data_path, image_dir, pred_dir)

   Function for predicting masks for all the input images.

   • checkpoint_path: Path where the architecture and weights of the model can be found.
   • data_path: Path of the directory where all the data are stored.
   • image_dir: Name of the directory where the images are stored.
   • pred_dir: Name of the directory where the predicted images will go.

Example Usages

Preprocessing:

from hotlib import preprocess

preprocess("/content/gdrive/MyDrive/Omdena/data", "inputs", "images", "masks")

Prediction:

from hotlib import predict

predict(
    "/content/gdrive/MyDrive/Omdena/checkpoints",
    "/content/gdrive/MyDrive/Omdena/data",
    "images",
    "predictions",
)