Sweden-based Mapillary, a premier member of NVIDIA Inception, uses deep learning to automate mapping. Their platform provides street-level mapping by stitching together images sourced from its community of individual contributors, companies, and governments. NVIDIA Inception is a startup accelerator.
Yesterday, the company announced the release of a new product, the Mapillary Street-Level Sequences dataset, a publicly available dataset that enables benchmarking and training of AI models for large-scale, lifelong place recognition.
Lifelong place recognition is the task of finding the most similar place of a query image in a database of geolocated images.
The Mapillary Street-Level Sequence dataset released this week is a collection of 1.6 million geographical image sequences bundled with metadata for training place-recognition algorithms.
Mapillary tools are made up of more than 1 billion images from over 190 countries. They use a variety of NVIDIA GPUs on the cloud and on-premises, including NVIDIA V100 GPUs for both training and inference. On the software side, they use NVIDIA DriveWorks SDK and TensorRT, NVIDIA’s programmable inference accelerator.
The company told NVIDIA that by adopting the NVIDIA TensorRT inference software stack in 2017, they were able to speed up segmentation algorithms by up to 27x when running on the Amazon Web Services cloud.
The Mapillary Street-Level Sequence Dataset is available as both a commercial edition and a research edition, the company stated in a post, Introducing the Mapillary Street-Level Sequences Dataset for Lifelong Place Recognition.
“We can’t wait to see the research results achieved with the Mapillary Street-Level Sequences Dataset in lifelong place recognition, and their applications for AR, robotics, and change detection,” the company said.