When: October 19th, 2025
Where: Honolulu, Hawai'i, ICCV 2025
Time: Morning
Preliminary Schedule
Opening [all organizers, 5 mins]
Invited Talk I: Richard Hartley [30 mins]
Invited Talk II: Eric Brachmann [30 mins]
Posters [70 mins]
Coffee Break [15 mins]
Invited Talk III: Gabriela Csurka [30 mins]
Invited Talk IV: Fredrik Kahl [30 mins]
Closing [all organizers, 10 mins]
CMT3 submission system: https://cmt3.research.microsoft.com/CALIPOSE2025
Paper submission deadline: June 28th, 2025 July 1st, 2025
Decisions: July 8th, 2025
Camera Ready Deadline: August 10th, 2025
Papers accepted to the workshop will be published in the ICCV 2025 workshop proceedings.
Paper submissions should use the ICCV 2025 paper template and follow the ICCV 2025 guidelines. In particular, paper submissions cannot be longer than 8 pages (excluding references).
Camera calibration is the problem of estimating the parameters of the projective mapping from 3D scene points to pixels in an image (or the inverse mapping from pixels to 3D rays in the scene). Camera pose estimation is the problem of determining the camera pose, i.e., the position and orientation, for a given image (or the relative pose between two images). (Nearly) all advanced 3D computer vision algorithms, including 3D reconstruction via NeRFs or 3D Gaussian splatting, as well as 3D scene understanding, require known camera calibrations and poses. The accuracy of the calibrations and the poses strongly impact the overall performance of these algorithms. Improving the accuracy, robustness, and efficiency of calibration and pose estimation methods can open doors to new applications, e.g., using advanced multi-camera systems in the space industry and for medical applications. However, most researchers do not concern themselves with accurate calibration. They either use calibrations and poses provided by a dataset, or use software such as COLMAP to compute them. The software is typically treated as a black box, with limited understanding about the accuracy of the results.
Calibration and pose estimation are also crucial problems in industry. In order to use cameras for downstream tasks such as tracking, recognition, segmentation, detection, navigation, etc., the cameras have to be calibrated. Moreover, industry often requires the calibration of special systems of cameras or other sensors, for which standard solutions are not directly applicable. While practically relevant, research on calibration and pose estimation of such multi-sensor systems is usually considered as not sufficiently novel to be published at the top-tier conferences.
Camera calibration and pose estimation have traditionally been based on solving (complex) equations arising from correspondences between pixels or pixels and 3D scene points. Although important open problems remain, these traditional approaches are becoming less and less popular. Recently, learning-based alternatives are gaining popularity, e.g., DUSt3R and MASt3R. However, these approaches do not yet reach the accuracy offered by traditional methods based on algebraic geometry. At the same time, it seems that the two communities working on classical algebraic methods, respectively, learning-based approaches, are rather disjoint.
The goals of the workshop are threefold: 1) Offer a forum for researchers working on classical approaches and learning-based approaches to meet and exchange ideas, with the aim of bringing both sub-communities closer together. Particularly interesting talking points are how classical and learning-based approaches can be combined to improve robustness and efficiency, and what type of applications become possible when having access to such advanced methods. To facilitate the discussion, the workshop provides four invited talks from well-known researchers, covering both classical and newer learning-based approaches. 2) Provide researchers and practitioners using camera calibration and pose estimation algorithms as black box tools the opportunity to learn about the current state-of-the-art and open problems in the field. This will enable them to better understand what existing approaches can and cannot do. 3) Provide researchers working on camera calibration and pose estimation methods a platform that, unlike the main conference, focuses on their research topics. To this end, our workshop invites the submission of novel, unpublished research in the form of paper submissions.
Topics of interest for the workshop include, but are not limited to:
Classical methods for camera calibration and camera (relative or absolute) pose estimation.
Learning-based methods for camera calibration and camera (relative or absolute) pose estimation.
Calibration of multi-camera and other multi-modal sensor systems.
Pose estimation from multiple/different modalities (images, depths, event cameras, etc.).
Combining classical and learning-based approaches for calibration and pose estimation.
Applications of camera calibration and camera pose estimation methods.
New theories on (learning-based) camera calibration and/or camera pose estimation.
Analysis of open challenges in the areas of calibration and/or pose estimation.
Datasets and benchmarks for related tasks.
The Microsoft CMT service was used for managing the peer-reviewing process for this conference. This service was provided for free by Microsoft and they bore all expenses, including costs for Azure cloud services as well as for software development and support.