NAVER在ECCV 2026获23篇视觉AI论文录用，4篇入选Spotlight

Team NAVER said on the 28th that 23 of its research papers have been accepted as regular papers at the European Conference on Computer Vision (ECCV) 2026, one of the world's most prestigious artificial intelligence conferences in the field of computer vision. The papers cover three-dimensional spatial recognition and reconstruction, image and video learning models, and robot navigation simulation. Now in its 19th edition, ECCV is regarded as the top authority in computer vision, including video and image processing. This year's conference runs from September 8 to 12 in Sweden.
NAVER's technology units — NAVER LABS, NAVER LABS Europe, NAVER AI Lab and NAVER Cloud — will present the accepted papers at the September conference. NAVER LABS Europe will introduce BLASt3R, a follow-up to DUSt3R, its three-dimensional spatial reconstruction technology. BLASt3R incorporates simultaneous localization and mapping (SLAM), a technique that lets a robot map its surroundings in real time as it moves, making three-dimensional reconstruction possible even in on-device environments. That approach sharply improved both speed and accuracy over existing models.
NAVER LABS' research on generating indoor three-dimensional spaces drew academic attention for reconstructing the layout of a room and the objects in it from a single 360-degree panoramic image. NAVER LABS also showcased technology that uses a vision language model (VLM) during autonomous driving to recognize objects it has not been trained on and place them in real time on an accurate bird's-eye view (BEV) map.
A study on the Seoul World Model, which NAVER AI Lab conducted in an industry-academic partnership with KAIST, was also accepted as a full paper at the conference, recognized for its potential use in map-based virtual exploration, digital twins and simulations for autonomous driving and robotics. The Seoul World Model draws on NAVER's vast trove of spatial data to generate long-range video that depicts the actual cityscape of Seoul.
NAVER Cloud also unveiled research that further advances AI's ability to understand images and video. Accepted papers included work on accurately recognizing the movement of objects within video and on adding text-based reasoning to images of spaces to make depth easier to grasp.
Four of the papers, including the research on 360-degree spatial reconstruction and the Seoul World Model, were selected for ECCV Spotlight, an honor given to roughly 1% to 2% of all submissions. The company sees the conference as fresh proof of Team NAVER's global competitiveness in spatial intelligence technology. A NAVER official said the company plans to keep pursuing aggressive research and development to secure next-generation AI and spatial intelligence technology and to apply it across its services, delivering technology experiences that users can readily feel in daily life.