Beyond Unidirectional Bias: Reciprocal Perspective Calibration in Scene Graph Generation

Scene Graph Generation (SGG) paradigms predominantly model relationships as static, unidirectional mappings ($s \to o$), effectively treating objects as passive recipients of actions. This formulation suffers from an inherent \textit{unidirectional bias}, violating the physical reality that visual interactions are intrinsically reciprocal. Consequently, existing models often fail to maintain logical self-consistency when the reasoning anchor shifts from the subject to the object. To rectify this cognitive deficiency, we establish the Mutual-Perspective Inverse Relations (MPIR) principle, positing that a robust visual representation must satisfy logical consistency across dual perspectives. Guided by this principle, we propose the \textbf{Reciprocal Perspective Calibration (RPC)} framework, a model-agnostic framework that operationalizes MPIR via a novel Adaptive Inverse-Relation Augmentation (AIRA) strategy. Furthermore, we introduce Hypernym-Guided Prompts (HGP) to bridge the gap between semantic context and computational efficiency in vision-language models, enabling precise modeling of inverse relations. Extensive experiments demonstrate that RPC not only achieves competitive performance on standard benchmarks but also significantly enhances the model's capability to understand inverse relations, as verified by a new inverse consistency evaluation protocol, demonstrating the cognitive robustness of our method.

Traditionally, vision models learn image relationships in one direction, like "a person riding a bicycle." But real-world interactions are reciprocal: the bicycle is simultaneously being ridden by the person. Current models struggle with this logical flip, leading to a flawed, one-sided understanding of visual scenes—a challenge we call the unidirectional bias. To fix this, our Mutual-Perspective Inverse Relations (MPIR) principle asserts that visual models must maintain logical consistency across dual perspectives. Guided by MPIR, we built the Reciprocal Perspective Calibration (RPC) framework to enforce these two-way constraints during training. Additionally, we use large language models as a semantic bridge, helping the model master the exact connection between an action and its structural reverse.