Accurate Positioning Even Without High-Definition Maps
Naver announced on the 8th that the 'Spatial Intelligence' technology of its technology research subsidiary Naver Labs ranked first in two categories at the '2024 European Conference on Computer Vision (ECCV)'.
ECCV is the world's most prestigious conference specializing in the field of computer vision, such as images and videos, where the latest artificial intelligence (AI) research results are presented and new technological topics are proposed. At ECCV 2024, Naver Labs challenged two categories: ▲Map-free visual re-localization ▲BOP Challenge, and won first place in both.
Since the 28th of last month, Naver Labs researchers have been presenting on MASt3R at ECCV 2024 held in Milan, Italy. [Photo by Naver]
The Map-free visual re-localization challenge tests how accurately localization can be performed even without precise maps. Since there are situations where maps cannot be created or prepared in advance, such as disasters or construction sites, the technology of 'localization without maps' is becoming increasingly important.
Naver Labs introduced an AI technology tool called MASt3R that reconstructs images in 3D for this challenge. MASt3R was recognized for its ability to achieve sufficiently accurate localization even without precise maps, ranking first among 12 participating teams including Google, Apple, and Meta. MASt3R is an upgraded version of DUSt3R, which is based on CroCo, a 3D vision foundation model (VFM) developed by Naver Labs Europe.
Naver Labs also took first place in the BOP (Benchmark for 6D Object Pose Estimation) challenge. The BOP challenge evaluates how accurately the 3D rotation and position of objects in images can be estimated. The technology model submitted by Naver Labs in this category was evaluated as the Best RGB-Only Method for the most accurate localization of objects using only red, green, and blue (RGB) images, and the Best Fast Method for the fastest processing, securing first place. This is the second consecutive year Naver Labs has won first place in the BOP challenge.
Seok Sang-ok, CEO of Naver Labs, said, "Spatial intelligence is a field we have been steadily preparing since the launch of Naver Labs by advancing robotics and autonomous driving technologies. We will further devote ourselves to research and development (R&D) to expand not only robotics, autonomous driving, and digital twins but also the supporting cloud infrastructure, core testbeds like 1784, and even new global milestones such as the Middle East."
Meanwhile, Naver Cloud also achieved the publication approval of 11 AI research papers at ECCV 2024. In particular, it attracted attention by presenting research proposing methods to more effectively refine large amounts of image and text data necessary for AI model creation to build high-quality training datasets, as well as studies introducing ways to improve the performance of image recognition models.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
