University of Leicester
Browse
Neural Belief Propagation for Scene Graph Generation.pdf (2.05 MB)

Neural Belief Propagation for Scene Graph Generation

Download (2.05 MB)
Version 2 2023-12-12, 12:51
Version 1 2023-12-11, 17:19
journal contribution
posted on 2023-12-12, 12:51 authored by Daqi Liu, Miroslaw Bober, Josef Kittler

Scene graph generation aims to interpret an input image by explicitly modelling the objects contained therein and their relationships. In existing methods the problem is predominantly solved by message passing neural network models. Unfortunately, in such models, the variational distributions generally ignore the structural dependencies among the output variables, and most of the scoring functions only consider pairwise dependencies. This can lead to inconsistent interpretations. In this article, we propose a novel neural belief propagation method seeking to replace the traditional mean field approximation with a structural Bethe approximation. To find a better bias-variance trade-off, higher-order dependencies among three or more output variables are also incorporated into the relevant scoring function. The proposed method achieves the state-of-the-art performance on various popular scene graph generation benchmarks. 

Funding

U.K. Defence Science and Technology Laboratory

10.13039/501100000266-Engineering and Physical Sciences Research Council

10.13039/100014036-Multidisciplinary University Research Initiative (Grant Number: EP/R018456/1)

History

Author affiliation

School of Psychology and Vision Science, University of Leicester

Version

  • VoR (Version of Record)

Published in

IEEE Transactions on Pattern Analysis and Machine Intelligence

Volume

45

Issue

8

Pagination

10161 - 10172

Publisher

Institute of Electrical and Electronics Engineers (IEEE)

issn

0162-8828

eissn

2160-9292

Copyright date

2023

Available date

2023-12-12

Language

en

Usage metrics

    University of Leicester Publications

    Categories

    Licence

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC