tl;dr: Machine learning model for generating VTT-compliant JSON from map images.
If you've ever downloaded a map file from the internet for your virtual role-playing session, you know the pain of identifying and quantifying gridlines by hand.
The goal of the Fantasy Maps project is to create a machine learning model that can identify gridlines on a map for you. Ultimately, the resulting model will be available for use as a C library, or even integrated into mobile and web applications for easy use!
In machine learning, your model is only as good as the quality and quantity of the data used to train it. To create a better model, you must provide more high-quality data.
This is where you come in: we need more training data! We accept computer-drawn maps with gridlines overlayed on top. Although we will accept just maps, we would really appreciate any VTT-compliant JSON metadata that goes along with the maps.
We accept PNG and JPG images.
The Fantasy Maps project uses Vertex AI, the flagship machine-learning platform from Google Cloud. Using Vertex AI, we are training AutoML object detection models to identify the gridlines on a map. The first step in training our model(s) are to create online prediction models to prove that the Fantasy Maps concept works. Ultimately, we will export the Vertex AI model(s) into a compact form that can be distributed along with a C library.
We have successfully created an online model that identifies gridlines!
Original image:
Predicted gridlines (output from model):
Our next step is to create an exportable version of this model to incorporate into a library. We need more training data to create a high-quality, exportable model.