A Causal Adjustment Module for Debiasing Scene Graph Generation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Liu, Li, Sun, Shuzhou, Zhi, Shuaifeng, Shi, Fan, Liu, Zhen, Heikkilä, Janne, Liu, Yongxiang
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909549410123776
author Liu, Li
Sun, Shuzhou
Zhi, Shuaifeng
Shi, Fan
Liu, Zhen
Heikkilä, Janne
Liu, Yongxiang
author_facet Liu, Li
Sun, Shuzhou
Zhi, Shuaifeng
Shi, Fan
Liu, Zhen
Heikkilä, Janne
Liu, Yongxiang
contents While recent debiasing methods for Scene Graph Generation (SGG) have shown impressive performance, these efforts often attribute model bias solely to the long-tail distribution of relationships, overlooking the more profound causes stemming from skewed object and object pair distributions. In this paper, we employ causal inference techniques to model the causality among these observed skewed distributions. Our insight lies in the ability of causal inference to capture the unobservable causal effects between complex distributions, which is crucial for tracing the roots of model bias. Specifically, we introduce the Mediator-based Causal Chain Model (MCCM), which, in addition to modeling causality among objects, object pairs, and relationships, incorporates mediator variables, i.e., cooccurrence distribution, for complementing the causality. Following this, we propose the Causal Adjustment Module (CAModule) to estimate the modeled causal structure, using variables from MCCM as inputs to produce a set of adjustment factors aimed at correcting biased model predictions. Moreover, our method enables the composition of zero-shot relationships, thereby enhancing the model's ability to recognize such relationships. Experiments conducted across various SGG backbones and popular benchmarks demonstrate that CAModule achieves state-of-the-art mean recall rates, with significant improvements also observed on the challenging zero-shot recall rate metric.
format Preprint
id arxiv_https___arxiv_org_abs_2503_17862
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Causal Adjustment Module for Debiasing Scene Graph Generation
Liu, Li
Sun, Shuzhou
Zhi, Shuaifeng
Shi, Fan
Liu, Zhen
Heikkilä, Janne
Liu, Yongxiang
Computer Vision and Pattern Recognition
Artificial Intelligence
While recent debiasing methods for Scene Graph Generation (SGG) have shown impressive performance, these efforts often attribute model bias solely to the long-tail distribution of relationships, overlooking the more profound causes stemming from skewed object and object pair distributions. In this paper, we employ causal inference techniques to model the causality among these observed skewed distributions. Our insight lies in the ability of causal inference to capture the unobservable causal effects between complex distributions, which is crucial for tracing the roots of model bias. Specifically, we introduce the Mediator-based Causal Chain Model (MCCM), which, in addition to modeling causality among objects, object pairs, and relationships, incorporates mediator variables, i.e., cooccurrence distribution, for complementing the causality. Following this, we propose the Causal Adjustment Module (CAModule) to estimate the modeled causal structure, using variables from MCCM as inputs to produce a set of adjustment factors aimed at correcting biased model predictions. Moreover, our method enables the composition of zero-shot relationships, thereby enhancing the model's ability to recognize such relationships. Experiments conducted across various SGG backbones and popular benchmarks demonstrate that CAModule achieves state-of-the-art mean recall rates, with significant improvements also observed on the challenging zero-shot recall rate metric.
title A Causal Adjustment Module for Debiasing Scene Graph Generation
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2503.17862