Invited Speakers

Workshop Information

• When: September 8th, 2026

• Time: PM (exact times to be released)

• Where: Malmö, Sweden

• Location: Malmö Arena and Malmömässan (room info TBD)

• Schedule: TBD

Workshop Description

3D mapping is the task of building 3D scene representations from a given set of images. Given a scene representation and a novel image, visual localization is the task of estimating the position and orientation from which the image was taken w.r.t. the scene representation. Both problems have a wide range of applications, including robotics, entertainment, and augmented reality. 

In practice, localization and mapping systems are often cloud-based: Given images sent to a server, the server builds a 3D map, or localizes the images w.r.t. a scene representation stored on the server. For example, a vacuum cleaning robot might build 3D models of its owner’s house in order to navigate the house. Due to limited memory and compute capabilities, the robot uses an external server to construct the 3D model. The same server is also used for localization, with the robot periodically sending images to the server. Obviously, anyone who gains access to the server can obtain private information, e.g., can learn about items in the house, security measures, etc. Privacy-preserving approaches to visual localization and mapping thus aim to obfuscate the images and / or the scene representation in order to prevent the disclosure of private information. Unfortunately, the field lacks clear definitions of what is considered private, preventing any actual proofs of privacy-preservation. This has led to works claiming to preserve privacy, only for later work to show that this is not the case, Our workshop aims to address this issue by:

•  Providing an extensive discussion on the privacy-localization accuracy tradeoff provided by existing privacy-preserving visual localization approaches. This discussion includes an evaluation framework for future work that provides a clearer definition of privacy.

• Giving an overview over the field and the current state-of-the-art through invited talks of experts in the field.

• Expanding the state-of-the-art through contributed papers.

Paper Submission

Deadline: August 7th

Notification: August 10th

Camera Ready Deadline: August 15th

OpenReview link: coming soon

Papers will be published in the ECCV 2026 workshop proceedings. Paper submissions must use the ECCV 2026 main conference template. Please note that all ECCV 2026 submission policies also apply to this workshop, including the 14 page limit, ensuring that the submission is blind and the double-submission rules. Submissions will be peer-reviewed and a decision will be made by the workshop organizers.

In addition to original work, we also solicit published / accepted papers covering the areas of interest of this workshop. These papers will not be published and not peer-reviewed, but authors will be invited to give a talk or present posters based on decisions made by the workshop organizers.

Call for Papers

The workshop focuses on privacy-preserving versions of the visual localization and 3D mapping problems. Visual localization and mapping are core computer vision problems. Their privacy-preserving variants are becoming increasingly more important with the growing availability of devices such as robots and AR glasses that constantly capture images and send them to cloud-based servers for further processing. The workshop covers all topics related to privacy-preserving localization and mapping.

Topics of interest for the workshop include, but are not limited to:

• Novel approaches for privacy-preserving localization and mapping

• (Learning-based) privacy-preserving image and 3D scene representations

• Privacy-preserving local image features

• Geometry obfuscation schemes for local features and 3D models

• Privacy guarantees for visual localization and mapping

• Camera pose estimation w.r.t. higher-level primitives, e.g., based on room outlines, object shapes, or object bounding boxes

• Localization and mapping using obfuscated images

• Scene coordinate regression for privacy-preserving localization and mapping

• Compact scene representations for visual localization

• Descriptor-free localization and mapping

• Privacy-preserving camera pose estimation

• Attacks on privacy-preserving representations for localization and mapping