Naver Recreates Indoor Spaces in 3D from a Single Photo; 23 Papers Accepted at Global Conference

[Financial News] Naver is accelerating the development of "spatial intelligence" technology, which enables artificial intelligence (AI) to see and understand real-world spaces. The company is working on technologies that can reconstruct an indoor space in 3D from a single photo, as well as systems that allow AI to learn urban environments and explore them virtually. Twenty-three related studies were accepted at a major global computer vision conference, underscoring the company’s technological strength.
Team Naver said on the 28th that 23 papers had been accepted for publication at the 2026 European Conference on Computer Vision (ECCV), one of the world’s leading computer vision conferences. Four of them were selected for the Spotlight track, which is awarded to only about 1% to 2% of all submissions.
This year marks the 19th edition of ECCV, which will be held in Sweden from Sept. 8 to 12. Technology teams from Team Naver, including NAVER LABS Corp., NAVER LABS Europe, Naver AI Lab and NAVER Cloud, will present the accepted papers.
The accepted papers cover a wide range of computer vision topics, from 3D spatial perception and reconstruction to image and video learning models and robot navigation simulation. In particular, a number of research achievements in 3D spatial reconstruction and spatial intelligence were accepted as regular papers.
NAVER LABS Europe will present Blaster, a follow-up study to its 3D vision model DUSt3R. Blaster is a model that incorporates simultaneous localization and mapping (SLAM), a technology that helps a robot determine its position while moving and build a map of its surroundings in real time. It improves processing speed and accuracy over the previous model so that 3D spaces can also be reconstructed in on-device environments.
NAVER LABS introduced a technology that can turn a single 360-degree panoramic image into a 3D representation of indoor structures and objects. The company will also present research that uses a vision-language model (VLM) in autonomous driving to recognize even objects it was not trained on and mark them on a real-time bird’s-eye-view (BEV) map.
A study called the Seoul World Model, jointly conducted by Naver AI Lab and KAIST, was also accepted. The model, trained on Seoul’s urban spaces, can be used for map-based virtual exploration, digital twin applications, and autonomous driving and robotics simulation.
NAVER Cloud will unveil research that enhances AI’s ability to understand images and videos. The work includes technology that identifies object movement in videos and technology that combines text-based reasoning with spatial images to make it easier to understand three-dimensional structures.
Among them, four papers, including the 360-degree spatial reconstruction technology InSpace and the Seoul World Model, were selected for the ECCV Spotlight track. Only about 1% to 2% of all submissions receive Spotlight recognition.
Team Naver plans to continue research into next-generation AI and spatial intelligence technologies, while also applying those achievements to real services so that users can experience them in everyday life.
[email protected] Choi Hye-rim Reporter