DGPE: A Dual-Stream Geometric-Aware Positional Encoding for Transformer-Based Object Detection

Accurately modeling complex spatial relationships is a pivotal challenge for Transformer-based object detectors, as existing models often rely on static positional encodings that fail to capture dynamic, real-world scene layouts. To address this issue, this work proposes a novel Dual-stream Geometri...

Popoln opis

Bibliografske podrobnosti

Principais autores:	Han-Cheng Hsiang, Zunliang Yan, Yong Liu
Format:	Artigo
Jezik:	Inglês
Izdano:	IEEE 2025-01-01
Serija:	IEEE Access
Teme:	Attention mechanism object detection positional encoding spatial reasoning Transformer
Online dostop:	https://ieeexplore.ieee.org/document/11278171/
Oznake:	Označite Brez oznak, prvi označite!