FairScene: Learning unbiased object interactions for indoor scene synthesis

Zhenyu Wu, Ziwei Wang, Shengyu Liu, Hao Luo, Jiwen Lu, Haibin Yan

Abstract: Highlights•We propose FairScene, a causal inference-based framework that reduces bias in indoor scene synthesis..•We mitigate the long-tail effect and confounding in object class distribution by using counterfactual samples and intervention.•Extensive experiments on the 3D-FRONT dataset validate the effectiveness of FairScene in object recognition and scene synthesis.