Automated surgical assistance systems rely on reliable instrument detection, yet current deep learning models remain sensitive to visual ambiguity. One overlooked factor contributing to this problem is perspective, which can hide or reveal critical discriminative features. We introduce a perspective-aware evaluation framework using synthetic data to analyze how perspective influences classification performance. Our results show that detection accuracy varies with viewpoint, identifying both optimal and ambiguous perspectives. These findings suggest that incorporating perspective-awareness may improve robustness in automated surgical systems.
Cerezo et al. (Wed,) studied this question.