Investigating Anisotropy in Visual Grounding under Controlled Counterfactual Perturbations

TL;DR

This paper investigates whether embedding anisotropy causes failures in visual grounding models under counterfactual perturbations, finding no significant correlation and suggesting the need to explore finer geometric properties.

Contribution

Introduces a similarity-controlled counterfactual caption generation protocol to analyze grounding behavior and tests it on two Transformer-based models with different embedding geometries.

Findings

No meaningful correlation between cosine similarity and approximation errors.

Embedding anisotropy alone does not explain counterfactual failures.

Robustness likely depends on finer-grained geometric properties of embeddings.

Abstract

Visual Grounding benchmarks assume that the object described by a referring expression is always present in the image, and grounding models are therefore rarely evaluated under semantically mismatched captions. In such cases, models frequently exhibit approximation behavior, producing a plausible bounding box that satisfies only part of the expression (\eg, preserving the original object while ignoring modified contextual cues). Because mismatched captions represent realistic edge cases, this behavior compromises reliability and raises concerns from an explainability perspective. Identifying its underlying causes is thus essential for improving model faithfulness and interpretability. Adopting a mechanistic interpretability viewpoint, this work examines whether embedding anisotropy contributes to counterfactual failures. A similarity-controlled counterfactual caption generation protocol is introduced to systematically perturb object or contextual components within predefined embedding similarity intervals, enabling a fine-grained analysis of grounding behavior as a function of alignment. Experiments on two Transformer-based models with markedly different embedding geometries (BERT-based TransVG and CLIP-based SwimVG) reveal no meaningful correlation between cosine similarity and approximation. These findings suggest that anisotropy alone does not account for counterfactual errors, and that robustness requires investigating finer-grained geometric properties of the embedding space.