Temporal feature fusion with deformable attention for multi-view 3D object detection

Open Webpage

Shaochen Zhang, Jinfu Yang, Tianhao Hao, Shuwen Wu, Mingai Li

Published: 2026, Last Modified: 29 Oct 2025Digit. Signal Process. 2026EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Designs DEConv to improve distant object detection by enhancing BEV edge features.•Proposes a temporal fusion network to reduce ghosting from redundant detection boxes.•Enables cross-branch BEV feature fusion with the DBDF module for better interaction.