ObjectRelator, a research developed at INSAIT, institute of Sofia University “St. Kliment Ohridski,” has been selected as a Highlight Paper at the upcoming ICCV 2025, one of the most prestigious conferences in artificial intelligence and computer vision. Only a small fraction of accepted papers are chosen for this recognition.
The project is a result of international collaboration with researchers from ETH Zurich, Fudan University, and the global technology company Ant Group, which serves over 1.3 billion users and 80 million merchants worldwide.
ObjectRelator represents a significant advance in developing AI systems capable of “seeing” the world in a human-like manner. The model can consistently understand objects across different perspectives—from what the human eye perceives (egocentric view) to what is captured by an external camera (exocentric view).
While humans naturally perceive and interpret objects across viewpoints, this remains a challenge for current AI systems. Most state-of-the-art vision models, such as SAM and SAM2, can only segment objects from a single image. ObjectRelator overcomes this limitation, enabling new applications in robotics, virtual and augmented reality, and intelligent agents capable of transferring knowledge seamlessly across viewpoints.
Congratulations to the entire team at INSAIT and our collaborators for this achievement: Yuqian Fu, Runze Wang, Bin Ren, Guolei Sun, Biao Gong, Yanwei Fu, Danda Pani Paudel, Xuanjing Huang, and Luc Van Gool.
Paper: arxiv.org/pdf/2411.19083
Project Page: objectrelator.ai (more interesting video demos)
Github Repo: https://github.com/insait-institute/ObjectRelator (all data, code, and models are released, check it out)