Exploring Scale Shift in Crowd Localization under the Context of Domain Generalization

2025-10-22

AI Summary

This work identifies and systematically addresses the scale shift problem in domain generalization (DG) for crowd localization—i.e., inconsistent head-scale distributions between training and test domains, leading to significant performance degradation. Through theoretical modeling and extensive experiments, we quantitatively demonstrate, for the first time, the detrimental impact of scale shift on 20 state-of-the-art DG algorithms, revealing it as a critical bottleneck for robust cross-domain localization. To tackle this, we introduce ScaleBench—the first DG benchmark specifically designed to evaluate scale-shift robustness—and propose Catto, a novel method that employs causal feature decomposition to disentangle scale-dependent and scale-invariant representations, coupled with an anisotropic scale adaptation mechanism. Evaluated across multiple cross-domain crowd localization benchmarks, Catto achieves substantial improvements in localization accuracy. Our work establishes scale-aware domain generalization as a new research direction in vision-based crowd analysis.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Intelligent Robots: Localization, Mapping, and Navigation

Application Category

• Security and Privacy: Large-scale security measurements
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• Search and Retrieval-Augmented AI: Vertical and domain-specific search

Abstract

Crowd localization plays a crucial role in visual scene understanding towards predicting each pedestrian location in a crowd, thus being applicable to various downstream tasks. However, existing approaches suffer from significant performance degradation due to discrepancies in head scale distributions (scale shift) between training and testing data, a challenge known as domain generalization (DG). This paper aims to comprehend the nature of scale shift within the context of domain generalization for crowd localization models. To this end, we address four critical questions: (i) How does scale shift influence crowd localization in a DG scenario? (ii) How can we quantify this influence? (iii) What causes this influence? (iv) How to mitigate the influence? Initially, we conduct a systematic examination of how crowd localization performance varies with different levels of scale shift. Then, we establish a benchmark, ScaleBench, and reproduce 20 advanced DG algorithms to quantify the influence. Through extensive experiments, we demonstrate the limitations of existing algorithms and underscore the importance and complexity of scale shift, a topic that remains insufficiently explored. To deepen our understanding, we provide a rigorous theoretical analysis on scale shift. Building on these insights, we further propose an effective algorithm called Causal Feature Decomposition and Anisotropic Processing (Catto) to mitigate the influence of scale shift in DG settings. Later, we also provide extensive analytical experiments, revealing four significant insights for future research. Our results emphasize the importance of this novel and applicable research direction, which we term Scale Shift Domain Generalization.

Problem

• Addressing scale shift in crowd localization domain generalization
• Quantifying scale shift impact on pedestrian detection performance
• Developing algorithms to mitigate domain generalization scale discrepancies

Innovation

• Proposed Causal Feature Decomposition algorithm
• Introduced Anisotropic Processing for scale shift
• Established ScaleBench benchmark for evaluation