CVF Computer Vision and Pattern Recognition Conference (CVPR2025)

POSTED: 09 Jul, 2025

In June, over 10,000 computer vision experts from around the world gathered in Nashville Tennessee for 2025 IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR 2025). Representing QUT Centre for Robotics, ARIAM Research Hub, Australian Centre for Robotics, and Abyss Solutions Ltd, Chamuditha Jayanga presented his paper ‘Multi-View Pose-Agnostic Change Localization with Zero Labels’. Co-authored with Jason Lai, Donald Dansereau, Niko Sünderhauf, and senior lead Dimity Miller, the work introduces a powerful new method for detecting scene changes—without labels or fixed camera poses.  

Chamuditha and the team present a novel label-free, pose-agnostic method for detecting scene changes using multi-view 3D Gaussian Splatting, outperforming existing approaches and enabling accurate change localisation from as few as five images – even at unseen viewpoints – alongside releasing a new real-world benchmark dataset.  

Check out the paper here: https://arxiv.org/pdf/2412.03911