2018 was a big year for the whole Mapillary community. We hit five million km (and flew right past it) and celebrated turning five years old. Because of our amazing community of both people and organizations, the Mapillary platform now hosts 422 million images covering 6.2 million km across the globe, with 36.2 billion objects detected in those images using computer vision. 2.2 million km worth of images were captured in 2018 alone.

We could not have done this without our dedicated community, partners, and customers. As always, we have happily remained committed to our cooperation with OpenStreetMap. We saw some great community mapping projects, and 29 people passed the milestone of 1 million images contributed. Our top contributor was r66 who uploaded more than 4.3 million (!) images in 2018 alone, bringing his all-time total to 6.6 million. We would also like to recognize bob3bob3 who mapped over 53,000 km of new roads this year.

This was a big year for our work with customers and partners as well. Our relationship with HERE and the Map Creator community was extended into a global partnership. The American Red Cross used Mapillary to collect imagery for their disaster response efforts. The state of Vermont brought five million images to the platform, which they are now able to integrate with their GIS tools. In Clovis, New Mexico, their one-man GIS shop was able to easily obtain traffic sign data—saving years of work and thousands of dollars.

Steven from Clovis prepares to capture street-level imagery by mounting action cameras on his vehicle

Throughout the year, we made several improvements to the Mapillary platform. We launched Mapillary for Organizations to help cities, agencies, and companies to work together as teams with shared access to their imagery and data. We released an update to our traffic sign recognition with increased support for more traffic sign classes, bringing the total amount of classes supported to 1,500! With this update, our detection rate is now 98%. The Mapillary web got a significant upgrade and loading times are now up to 80% faster than before.

To make contributing images more accessible, our capture and upload tools got some upgrades. This year we released our first software development kits. These SDKs allow anyone to build a street-level imagery capture component into their own app. We also updated the Mapillary command line tools to make uploading images faster and smoother than ever before and on that same note, today we made the last announcement of 2018 when we launched our first Desktop Uploader.

2018 saw more innovative updates from the Computer Vision team. The Mapillary Vistas Dataset now contains more than 150 object classes, up from 100. It remains the world’s most diverse publicly available dataset of street-level imagery for training autonomous systems and algorithms. We announced our work with Amazon Rekognition to help US cities put an end to their parking problems. We also introduced Undistortion, which uses computer vision to improve the viewing experience of Mapillary images and gives a better 3D representation. The team also developed an updated privacy blurring model that is able to detect 99.9% of identifiable faces and license plates.

Our Research team has also been busy in 2018. Using an algorithm trained on Vistas to build their models, our team won the CVPR Semantic Segmentation Challenge and the ECCV Semantic Segmentation Challenge for Autonomous Navigation in Unstructured Environments. We also co-hosted a recognition challenge workshop with COCO, co-located with ECCV. Finally, the team published and announced InPlace-ABN, a new way to slash up to 50% of GPU memory required when training deep neural network models.

The winners of our workshop with COCO were announced to a packed house at ECCV

We are looking forward to another exciting year, and want to say a big thank you to our entire community and all our supporters out there. You continue to amaze us on a daily basis, and it’s a joy to be on this journey with you. Happy 2019!

/Jan Erik and the Mapillary team