Scene reconstruction from unposed 2D images is a long-standing, multi-disciplinary research area that spans computer vision, graphics, and photogrammetry. There exist many significant implications from this area that can benefit society, e.g., in autonomous navigation, augmented reality, smart cities, disaster relief planning, etc. On a high level, the main problem can be described into two parts: camera calibration and dense 3D reconstruction. In recent years, these two areas have seen significant interest and progress, particular with dense reconstruction.

The goal of this full-day WACV2025 workshop is to to examine the progress of 3D scene reconstruction/rendering, its underlying assumptions, and the challenges that lie ahead in unconstrained and large scale scenarios. To this end, ULTRRA will release a series of datasets with carefully measured ground-truth. These datasets are crafted to emphasize the challenges in camera calibration and 3D reconstruction/rendering. A challenge will be held to quantitatively and qualitatively evaluate current and novel SoTA’s performances.

• A real-time visualization of some of the scenes can be access through the following barcode (password: wriva), courtesy to Accenture Federal Services and Deva Ramanan's group at CMU! 

• The ULTRRA Challenge dataset is out at https://ieee-dataport.org/competitions/ultrra-challenge-2025!

• We accept either 4-page extended abstracts or 8-page full papers submissions, excluding reference. The workshop papers are non-archival and we welcome submissions that were already submitted/accepted to other venues or the WACV main coference. All submissions should follow the WACV 2025 author guidelines.

  • Submission Portal: CMT

  • Paper Submission Deadline: 12/16/2024

  • Notification to Authors: 12/23/2024

  • Camera-ready submission: 01/05/2024

• Accepted papers will be invited for poster/oral presentation and will be displayed on the workshop website.

The ULTRRA challenge evaluates current and novel state of the art view synthesis methods for unposed cameras. Challenge datasets will emphasize real-world considerations, such as image sparsity, variety of camera models, and unconstrained acquisition in real-world environments. 

• Tentative Schedule

  • Development dataset release: 11/8/2024 (Available Now @ https://ieee-dataport.org/competitions/ultrra-challenge-2025!)

  • Challenge dataset release: 1/10/2025 (Available Now @ https://ieee-dataport.org/competitions/ultrra-challenge-2025!)

  • Submission period: 11/1/2024 to 2/14/2025 (tentative) 

  • Winners presentation:  2/28/2025

• Dataset

  • Images collected for the IARPA WRIVA program will be publicly released and made available for use in this public challenge and more broadly to encourage research in view synthesis methods for real-world environments and heterogeneous cameras. Datasets include images collected from mobile phones and other ground-level cameras, security cameras, and airborne cameras. Each camera is calibrated using structure from motion constrained by RTK-corrected GPS coordinates, with accuracies measured in centimeters, for either camera locations or ground control points, depending on the camera. Cameras are geolocated to enable reliable evaluation. Images used for final evaluations will be sequestered. 

  • The data has been collected and is in public release review. The development datasets have already been constructed. Challenge datasets are in work and will be like the development datasets but using sequestered images for evaluation. We anticipate posting the development datasets on IEEE DataPort November 1st 2024 and the challenge datasets January 10th 2025. 

• Submission Evaluation

  • The competition will be hosted on CodaBench (https://www.codabench.org/) or an equivalent platform. Challenge tracks will explore multi-camera pose estimation and photo-realistic novel view synthesis. Challenge datasets will include input images from a variety of ground-level, security, and airborne cameras. Camera poses will be evaluated by comparing relative camera locations, with contestant coordinate frames aligned to reference coordinates using Procrustes analysis. Sequestered images and contestant rendered images will be compared using the DreamSim image similarity metric (https://dreamsim-nights.github.io/) to establish view synthesis scores.

Coffee break 10:30 - 10:45

Workshop Spotlight Presentation 11:30 - 12:35 (Schedule)

Workshop Poster Session 12:35 - 14:00

Coffee break 15:30 - 15:45

Workshop Challenge Session 15:45 - 16:40

This work was supported by the Intelligence Advanced Research Projects Activity (IARPA) contract no. 2020-20081800401. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright annotation thereon. Disclaimer: The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of IARPA or the U.S. Government. 

This work was supported by the Intelligence Advanced Research Projects Activity (IARPA) via Department of Interior/ Interior Business Center (DOI/IBC) contract number 140D0423C0076. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright annotation thereon. Disclaimer: The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of IARPA, DOI/IBC, or the U.S. Government