TEAM NAVER Demonstrates Global Expertise in Spatial Intelligence and AI at Leading Vision Conference
TEAM NAVER Demonstrates Global Expertise in Spatial Intelligence and AI at Leading Vision Conference
TEAM NAVER Demonstrates Global Expertise in Spatial Intelligence and AI at Leading Vision Conference
- TEAM NAVER has 14 papers accepted at CVPR 2025, with a total of 151 papers published at the world’s top three computer vision conferences over 5 years
- Research achievements in various fields, including 3D spatial recognition and reconstruction, image learning models, and robotics
- Over 450 papers presented at top AI conferences in NLP, vision, speech, and machine learning, with more than 47,000 citations following proactive R&D investment
March 25, 2025
A total of 14 research papers from various technology organizations within TEAM NAVER, including NAVER LABS, NAVER LABS Europe, and NAVER Cloud, have been accepted for publication at Computer Vision and Pattern Recognition Conference (CVPR) 2025, one of the world’s most prestigious conferences in the field of computer vision. This achievement demonstrates the competitiveness of NAVER’s spatial intelligence and vision AI technologies. Founded in 1983, CVPR is an academic conference co-hosted by the Institute of Electrical and Electronics Engineers (IEEE), the world’s largest technical professional organization, and the Computer Vision Foundation (CVF). It is recognized as a leading conference in the field of vision AI and one of the most influential conferences in computer science.
* Ranked second overall in all scientific disciplines, following Nature, and first in the field of Engineering and Computer Science according to Google Scholar’s conference and journal rankings.
With this achievement, NAVER has now published 151 regular papers over the past 5 years (from 2020 to March 2025) in the world’s top three global computer vision conferences: CVPR, the European Conference on Computer Vision (ECCV), and the International Conference on Computer Vision (ICCV). During this period, NAVER has consistently published double-digit numbers of papers annually, solidifying its position as a world-class AI technology company.
Research achievements in various fields, including 3D spatial recognition and reconstruction, image learning models, and robotics
At CVPR 2025, TEAM NAVER will first present follow-up research results on the 3D reconstruction AI tool “DUSt3R,” which NAVER LABS Europe unveiled last year, garnering significant interest from global tech giants. Based on DUSt3R, AI technology that enables easy 3D reconstruction from just one or two photos, new AI models have been introduced. These include “MUSt3R,” which offers more accurate 3D reconstruction from multiple images, and “Pow3R,” which enhances inference capabilities by integrating various camera and scene information. In addition, several spatial intelligence papers have been accepted at the conference, including a NAVER LABS paper proposing a technology for swiftly and accurately detecting the position and orientation of previously untrained objects (“Co-op: Correspondence-based Novel Object Pose Estimation”).
Research on image learning models, which have gained attention with the rise of multimodal AI technology, has also been accepted. NAVER Cloud proposed an efficient learning methodology to address training instability issues in AI image learning through the use of the “masking” technique (“Masking meets Supervision: A Strong Learning Alliance”). Meanwhile, NAVER LABS Europe introduced a technology that precisely distinguishes new objects in images without additional training by utilizing a Vision-and-Language Model (“LPOSS: Label Propagation Over Patches and Pixels for Open-vocabulary Semantic Segmentation”).
Achievements in robotics research also stood out. A study on an autonomous driving system, where a fast-moving robot efficiently finds its path using an “End-to-End” learning approach that covers the entire process from visual input to action output, was accepted (“Reasoning in visual navigation of end-to-end trained agents: a dynamical systems approach”), along with a study proposing methods to enhance the creativity of image-generation AI models (“Enhancing Creative Generation on Stable Diffusion-based Models”).
Over 450 papers and 47,000 citations in top-tier AI conferences across fields such as NLP, vision, speech, and machine learning
TEAM NAVER has achieved outstanding research results not only in computer vision but also in top-tier AI conferences across fields such as natural language processing (NLP), speech, and machine learning. To date, the team has published over 450 research papers, with about 47,000 citations, one of the key indicators of research impact.
NAVER announced that it will continue to invest actively in research and development (R&D) to secure advanced technologies. Under the “On-Service AI” strategy, NAVER also plans to integrate key technologies it has successfully internalized, such as generative AI and spatial intelligence, into major services to provide users with a differentiated service experience. </End>
[Reference] List of TEAM NAVER Papers Accepted for CVPR 2025
1. MUSt3R: Multi-view Network for Stereo 3D Reconstruction
2. Pow3R: Empowering Unconstrained 3D Reconstruction with Camera and Scene Priors.
3. Reasoning in visual navigation of end-to-end trained agents: a dynamical systems approach
4. LPOSS: Label Propagation Over Patches and Pixels for Open-vocabulary Semantic Segmentation
5. Heterogeneous Teacher Distillation
6. Gaussian Splatting Feature Fields for (Privacy-Preserving) Visual Localization
7. Layered motion fusion: Lifting motion segmentation to 3D in egocentric videos
8. Co-op: Correspondence-based Novel Object Pose Estimation
9. EDM: Equirectangular Projection-Oriented Dense Kernelized Feature Matching
10. Masking meets Supervision: A Strong Learning Alliance
11. Enhancing Creative Generation on Stable Diffusion-based Models
12. ControlFace: Harnessing Facial Parametric Control for Face Rigging
13. CoCoGaussian: Leveraging Circle of Confusion for Gaussian Splatting from Defocused Images
14. Nearly Zero-Cost Protection Against Mimicry by Personalized Diffusion Models
