As far as history’s concerned, 2020 was a remarkable year. It’ll be a while before the implications are understood, but it’s increasingly evident that profound changes are underway. If we narrow in on Mapillary, 2020 was a big year for us as well. New features were released, computer vision research and datasets made available, and the mapping community continued to wow us with their hard work and creativity. Amidst all this, we joined Facebook where we have focused even more heavily on how Mapillary can improve maps.

Below we’d like to recap some of the highlights of the year across the platform, Mapillary Research, and the community.

Mapillary platform and research

OpenStreetMap

In 2020, we were able to start delivering on some of our longer term goals for computer vision assisted map editing in OpenStreetMap. There were two notable releases here:

1. Map features were made fully available for both iD Editor and JOSM, allowing map editors to view detections such as utility poles and crosswalks while editing the map.
2. Smart Edits builds upon the map features release, making it easier to convert a Mapillary detection to a node in OpenStreetMap.

Integrations

Following popular demand and feedback from partners, Mapillary is now integrated in two new applications.

1. ArcGIS Urban is a tool from Esri used by urban planners, local governments, engineers, students, or anyone else wanting to visualise environments in 3D. Incorporating street-level imagery into this tool made a lot of sense because 3D building data can now be viewed alongside a recent photo of that building.
2. Geocortex is a GIS tool to help visualise datasets that have a spatial component. It’s a prominent application for local governments and engineering firms and a few have been asking how they can view street-level imagery directly in Geocortex. This new integration allows them to work with their data with the context that street-level imagery provides.

Mapillary viewer alongside ArcGIS Urban.

Computer Vision

It was a busy year for the computer vision team with datasets released, new automatic detections, and research papers presented.

• Datasets released:
  • Mapillary Street-level Sequences (MSLS) dataset which helps computers to learn where an image has been taken.
  • Mapillary Planet Scale Depth (MPSD) dataset which helps computers to infer 3D information from a single image.
• Scene classes: These machine generated map features highlight distinct types of infrastructure such gas stations and parking lots.
• Accepted papers: 7 papers were accepted and presented at CVPR and ECCV, two of the most prominent computer vision conferences.

Privacy

To further enhance privacy, we now blur faces and license plates as soon as the images hit our servers. The original images are deleted as soon as the blurring has been completed.

Community

#Map2020 returns

#Map2020 is a collaboration between Mapillary, the Humanitarian OpenStreetMap Team, and local mapping communities. Participating groups find ways to apply street-level imagery and data to local challenges. This year we focused on urban mobility and the many ways citizens navigate their cities. Natália Arruda and her project SIGenBici in Medellín, Colombia was chosen as the project of the year and presented at the 2020 HOT Summit.

YouthMappers of Makerere University taking part in #Map2020. This photo and others from @GeoYouthmappers.

Mapping Sierra Leone’s power infrastructure

YouthMappers in Sierra Leone have been driving a project to map the country’s electrical grid. This will help the roll-out of mini-grids which are critical for rural electrification. Street-level imagery, map data, Pic4Review, XENDEE, OpenDSS, and other tools. The project brings together a wide array of partners including Arizona State University, USAID, and Sierra Leone’s Ministry of Energy.

Community verifications and their impact on recall and precision

Community input has played an essential role in improving the precision and recall of our machine generated object detections. In 2020, Gerhard revealed the ways over 590,000 verifications were used to improve our deep neural networks. The full results can be viewed in his blog post.

Joining Facebook

In June, 2020, Mapillary joined the team at Facebook where we have continued our mission to map the world with street-level imagery. It has been a busy and exciting time getting to know our new team and working to make Mapillary more useful to our community of users.

One of the most exciting aspects of this change has been the chance to open up Mapillary imagery and map data for commercial use. Mapillary imagery and data has always been available for non-commercial use, but commercial users had to purchase access. Making data available for both use cases encourages more people to contribute imagery which in turn means a wider variety of map data in more places around the world.

2021 and beyond

2020 was a tough year for many, but there were also positives to take away. We are grateful to all of you for the important role you played in the year that was.

We hope you enjoyed this snapshot of our 2020 highlights. What were your mapping highlights in 2020? What are you excited about as 2021 marches on?