Niantic organises the first #MapFreeReloc workshop and challenge at #ECCV2024 European Conference on Computer Vision #ECCV2026! We invite the submission of extended abstracts and papers. The challenge features two tracks, and includes a multi-frame version and $6000 in prizes. Page:

Map-free relocalisation was proposed in an eponymic #ECCV2022 paper. How accurately can we estimate the relative pose between two images, a reference and a query, including metric scale. #MapFreeReloc enables instant AR everywhere, without having to build visual maps upfront.

The task comes with a substaintial dataset that includes challenging image pairs with little or no visual overlap. Reasoning about object shapes is necessary to solve some of the cases. Methods need to provide a robust confidence associated with their pose estimates.

Since our public leaderboard went online, a little over one year ago, we have seen tremendous progress. So much in fact, that we are able to slash our evaluation threshold from 90px to 45px, requiring significantly more accurate estimates for the challenge.

Based on community feedback, we added a second track for the challenge: The reference is still a single image but the query consists of a burst of frames with tracking poses, corresponding to a small camera motion. We are curious to see how much it helps!

We invite papers and extended abstracts, even if you do not participate in the challenge. #MapFreeReloc is associated with a broad range of topics, such as visual relocalisation, feature matching, depth estimation, pose regression, uncertainty and confidence, and SfM.

Sign up for updates: The workshop is organized by @AMonszpart with the help of @FilipeGGaspar, Guillermo Garcia-Hernando (@gui_ggh), @axelbarrosotw, Daniyar Turmukhambetov (@dantkz), Gabriel J. Brostow, Victor Adrian Prisacariu (@viprad), and myself.

@eccvconf Niceee!

@eccvconf How did you make these beautiful visualizations? Blender or Rerun? 🤔

@eccvconf Some custom PyRender code. If I find the time to clean it up, we might release it. It's quite useful to inspect results.