2026-09-02 –, Ran1
Mapillary hosts over 2 billion street-level images as open data. Their quality and capture conditions vary. This presentation introduces filtering methods to remove low-quality or irrelevant images, enabling more effective use of Mapillary imagery.
Mapillary is an open data platform where users upload and share street-level imagery. It contains over 2 billion street-level images from around the world. These images are available as open data (CC BY-SA 4.0) and are widely used in various fields, such as OpenStreetMap mapping and urban environment assessment.
Image quality on Mapillary can vary due to the diversity of contributors and capturing devices. The images vary in viewpoint, resolution, weather conditions, and time of capture, and some contain motion blur or are out of focus. Therefore, users need to not only retrieve images from target regions, but also filter out low-quality or irrelevant images using metadata and image analysis techniques, such as semantic segmentation and vision-language models (VLMs).
In this presentation, I introduce methods for filtering images and share practical experiences from applying these methods. This session provides hints for using Mapillary imagery more effectively.
Banno is a PhD student at the University of Tsukuba. His research focuses on urban streetscape analysis using street-level imagery. He has an MSc in biology and worked as a software engineer until March 2026.