One paper from NTT Laboratories have been accepted for publication for ICCV 2025

One paper authored by NTT Laboratories has been accepted for publication at ICCV 2025 (IEEE International Conference on Computer Vision)(https://iccv.thecvf.com/), the flagship conference on computer vision to be held in Honolulu, Hawaii, from 19th to 23rd October, 2025. ICCV is known as one of the most competitive conferences in computer vision. The acceptance rate of ICCV 2025 is 24.0% (2,698 out of 11,239).

HI: NTT Human Informatics Laboratories

MVTrajecter: Multi-View Pedestrian Tracking with Trajectory Motion Cost and Trajectory Appearance Cost

Taiga Yamane (HI), Ryo Masumura (HI), Satoshi Suzuki (HI), Shota Orihashi (HI)

Multi-view pedestrian tracking aims to track pedestrians in the form of a bird's eye view from a multi-view video. In this task, end-to-end deep learning methods, which detect and associate pedestrians in one model, have made significant progress. However, previous methods rely solely on information between two frames to associate pedestrians, leading to degradation in tracking performance due to missed detections of pedestrians. This study proposes a novel end-to-end method for more robust pedestrian tracking by utilizing information from more frames for pedestrian association.