HiLo: Exploiting High Low Frequency Relations for Unbiased Panoptic Scene Graph Generation

被引:3
|
作者
Zhou, Zijian [1 ]
Shi, Miaojing [2 ]
Caesar, Holger [3 ]
机构
[1] Kings Coll London, Dept Informat, London, England
[2] Tongji Univ, Coll Elect & Informat Engn, Shanghai, Peoples R China
[3] Delft Univ Technol, Intelligent Vehicles Lab, Delft, Netherlands
关键词
D O I
10.1109/ICCV51070.2023.01978
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Panoptic Scene Graph generation (PSG) is a recently proposed task in image scene understanding that aims to segment the image and extract triplets of subjects, objects and their relations to build a scene graph. This task is particularly challenging for two reasons. First, it suffers from a long-tail problem in its relation categories, making naive biased methods more inclined to high-frequency relations. Existing unbiased methods tackle the long-tail problem by data/loss rebalancing to favor low-frequency relations. Second, a subject-object pair can have two or more semantically overlapping relations. While existing methods favor one over the other, our proposed HiLo framework lets different network branches specialize on low and high frequency relations, enforce their consistency and fuse the results. To the best of our knowledge we are the first to propose an explicitly unbiased PSG method. In extensive experiments we show that our HiLo framework achieves state-of-the-art results on the PSG task. We also apply our method to the Scene Graph Generation task that predicts boxes instead of masks and see improvements over all baseline methods. Code is available at https://github.com/franciszzj/HiLo.
引用
收藏
页码:21580 / 21591
页数:12
相关论文
共 50 条
  • [41] PCPL: Predicate-Correlation Perception Learning for Unbiased Scene Graph Generation
    Yan, Shaotian
    Shen, Chen
    Jin, Zhongming
    Huang, Jianqiang
    Jiang, Rongxin
    Chen, Yaowu
    Hua, Xian-Sheng
    MM '20: PROCEEDINGS OF THE 28TH ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA, 2020, : 265 - 273
  • [42] Joint Modeling of Visual Objects and Relations for Scene Graph Generation
    Xu, Minghao
    Qu, Meng
    Ni, Bingbing
    Tang, Jian
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 34 (NEURIPS 2021), 2021, 34
  • [43] Triple Correlations-Guided Label Supplementation for Unbiased Video Scene Graph Generation
    Wang, Wenqing
    Gao, Kaifeng
    Luo, Yawei
    Jiang, Tao
    Gao, Fei
    Shao, Jian
    Sun, Jianwen
    Xiao, Jun
    PROCEEDINGS OF THE 31ST ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA, MM 2023, 2023, : 5153 - 5163
  • [44] A New Training Data Organization Form and Training Mode for Unbiased Scene Graph Generation
    Xu, Hongbo
    Wang, Lichun
    Xu, Kai
    Fu, Fangyu
    Yin, Baocai
    Huang, Qingming
    IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, 2024, 34 (07) : 5295 - 5305
  • [45] Stacked Hybrid-Attention and Group Collaborative Learning for Unbiased Scene Graph Generation
    Dong, Xingning
    Gan, Tian
    Song, Xuemeng
    Wu, Jianlong
    Cheng, Yuan
    Nie, Liqiang
    2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2022), 2022, : 19405 - 19414
  • [46] Skew Class-Balanced Re-Weighting for Unbiased Scene Graph Generation
    Kang, Haeyong
    Yoo, Chang D.
    MACHINE LEARNING AND KNOWLEDGE EXTRACTION, 2023, 5 (01): : 287 - 303
  • [47] State-Aware Compositional Learning Toward Unbiased Training for Scene Graph Generation
    He, Tao
    Gao, Lianli
    Song, Jingkuan
    Li, Yuan-Fang
    IEEE TRANSACTIONS ON IMAGE PROCESSING, 2023, 32 : 43 - 56
  • [48] Not All Relations are Equal: Mining Informative Labels for Scene Graph Generation
    Goel, Arushi
    Fernando, Basura
    Keller, Frank
    Bilen, Hakan
    2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2022), 2022, : 15575 - 15585
  • [49] From Easy to Hard: Learning Curricular Shape-Aware Features for Robust Panoptic Scene Graph Generation
    Shi, Hanrong
    Li, Lin
    Xiao, Jun
    Zhuang, Yueting
    Chen, Long
    INTERNATIONAL JOURNAL OF COMPUTER VISION, 2025, 133 (01) : 489 - 508
  • [50] DBiased-P: Dual-Biased Predicate Predictor for Unbiased Scene Graph Generation
    Han, Xianjing
    Song, Xuemeng
    Dong, Xingning
    Wei, Yinwei
    Liu, Meng
    Nie, Liqiang
    IEEE TRANSACTIONS ON MULTIMEDIA, 2023, 25 : 5319 - 5329