Logical Bias Learning for Object Relation Prediction

TL;DR

This paper introduces a causal inference-based strategy for object relation prediction in scene graph generation, addressing data bias issues and improving performance on the VG-150 dataset.

Contribution

It proposes a novel causal inference approach for relation prediction and an object enhancement module to validate its effectiveness.

Findings

Improved relation prediction accuracy on VG-150 dataset

Effective mitigation of data bias in scene graph generation

Demonstrated potential for foundation models in decision-making

Abstract

Scene graph generation (SGG) aims to automatically map an image into a semantic structural graph for better scene understanding. It has attracted significant attention for its ability to provide object and relation information, enabling graph reasoning for downstream tasks. However, it faces severe limitations in practice due to the biased data and training method. In this paper, we present a more rational and effective strategy based on causal inference for object relation prediction. To further evaluate the superiority of our strategy, we propose an object enhancement module to conduct ablation studies. Experimental results on the Visual Gnome 150 (VG-150) dataset demonstrate the effectiveness of our proposed method. These contributions can provide great potential for foundation models for decision-making.