Tackling the Unannotated: Scene Graph Generation with Bias-Reduced Models

TL;DR

This paper introduces a bias-reduction training scheme for scene graph generation that improves mean class-wise recall significantly while maintaining competitive overall recall, addressing annotation bias issues in visual datasets.

Contribution

A novel training approach using dual relation classifiers to mitigate annotation bias in scene graph generation models, applicable to existing architectures.

Findings

Significant improvements in mean recall (6.6% to 20.4%) across models.

Maintained or slightly improved overall recall (−2.4% to 0.3%).

Applicable to most existing SGG models with straightforward implementation.

Abstract

Predicting a scene graph that captures visual entities and their interactions in an image has been considered a crucial step towards full scene comprehension. Recent scene graph generation (SGG) models have shown their capability of capturing the most frequent relations among visual entities. However, the state-of-the-art results are still far from satisfactory, e.g. models can obtain 31% in overall recall R@100, whereas the likewise important mean class-wise recall mR@100 is only around 8% on Visual Genome (VG). The discrepancy between R and mR results urges to shift the focus from pursuing a high R to a high mR with a still competitive R. We suspect that the observed discrepancy stems from both the annotation bias and sparse annotations in VG, in which many visual entity pairs are either not annotated at all or only with a single relation when multiple ones could be valid. To address this particular issue, we propose a novel SGG training scheme that capitalizes on self-learned knowledge. It involves two relation classifiers, one offering a less biased setting for the other to base on. The proposed scheme can be applied to most of the existing SGG models and is straightforward to implement. We observe significant relative improvements in mR (between +6.6% and +20.4%) and competitive or better R (between -2.4% and 0.3%) across all standard SGG tasks.