Deep Pavements is a framework to leverage open-source tools to produce data about pathway pavements using CC-compatible street-level Imagery. As a project highlight, the pavement classes are compliant with the OpenStreetMap surface=* tag values.
The main products that are meant to be created are with the provided tools are:
- I Regionally optimized Deep Learning models to do segmentation of road pavements and objects that may interfere with them.
- Datasets to train semantic segmentation models.
- A surface-patch dataset to train image classification models.
The modules that made up the project are each one in a separate repository:
- Sample Picker: to generate samples from randomly selected images from Mapillary.
- Sample Labeler: To manually check the samples.
- Surface Patch Dataset: A manually curated dataset of surface patches.
- Model Trainer: To train semantic segmentation models.
A module "Data Processer" is also planned. This module will be used to generate the surface patches for an specified location.
[1] : The logo was produced using Microsoft Copilot, we plan to replace it with in the future with a CC one.