Bias-aware multimodal object detection through structured latent graph decoding

Ziyu Li

doi:10.1016/j.neucom.2026.134897

Bias-aware multimodal object detection through structured latent graph decoding

2026-08-25

0

PRE

AI

DOI：10.1016/j.neucom.2026.134897

Original

Original request for help

Save

Abstract

En 中文

• A novel framework GraDBridge is designed for robust multispectral object detection. • GeoBridge Evolution (GBE) builds branch-specific Geo Reference Kernels for latent graph interaction. • Latent Barycentric Positional Bias (LBPB) reduces latent-to-spatial decoding ambiguity via barycentric priors. • The proposed GraDBridge achieves superior accuracy on M3FD and FLIR detection benchmarks.

Bias-aware multimodal object detection through structured latent graph decoding

Abstract

Journal

Organization