mapyllary

Unofficial API to request image and segmentation from Mapillary.
This package is hosted at https://pypi.org/project/mapyllary/
Note: The OSM-related functionality of this package is imported from overpass https://github.com/mvexel/overpass-api-python-wrapper

Install

This package requires python version >= 3.5

pip install mapyllary

Note that installation will take a while due to dependency on ML & CV libraries

How to obtain a Mapillary client ID

Requests to Mapillary server must be identified by a client id, and you can obtain one for free (as of now) by registering an account here. https://www.mapillary.com/ Then, go to your Dashboard > Developer to generate your client ID.

Test run

You can run the main.py as a command line tool to download images from Mapillary around a certain GPS coordinates, or a collection of locations obtained by OpenStreetMap filters.

Instructions

1. git clone https://github.com/boycetsang/mapyllary.git
2. Download images from a GPS coorindate (long, lat): python main.py -c <your_client_id> -p -122.872700 45.543663
3. Download images from a filter (OSM style; e.g. all the highway nodes that are motorway_junction within Washington state): python main.py -c <your_client_id> -n highway motorway_junction Seattle
4. Bonus: Applying a segmentation model to downloaded images python main.py -c <your_client_id> -n highway motorway_junction "North Plains" -m model/enet-model.net -l model/enet-label.txt -w 1024 -t 512
5. All the resultant image files are located at "resources" directory under current directory.

Comparison between original image, Mapillary segmentation data and a LaneNet Model

Using the API

Here is a colab notebook that illustrate how the API can be used to request images and segmentation data from Mapillary.

https://colab.research.google.com/drive/15V9A7Z7oiOoZlNB4_9upXnPauOxPvmif?usp=sharing