Exploring Scale Shift in Crowd Localization under the Context of Domain Generalization
Not on the map. Why?
Abstract (arXiv)
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.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 112 of its 160 references that are in the map. First posted on arXiv 22 Oct 2025.
Builds on
112 of 160 references.
References · 112
- 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
- 2014Microsoft COCO: Common Objects in Contextcited by 17,608
- 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,974
- 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,403
- 2025Qwen2.5-VL Technical Reportcited by 5,587
- 2017mixup: Beyond Empirical Risk Minimizationcited by 5,183
- 2016Feature Pyramid Networks for Object Detectioncited by 5,036
- 2022LAION-5B: An open large-scale dataset for training next generation image-text modelscited by 4,152
- 2015Domain-Adversarial Training of Neural Networkscited by 3,044
- 2019Invariant Risk Minimizationcited by 1,854
- 2019Deep High-Resolution Representation Learning for Visual Recognitioncited by 1,420
- 2020Sharpness-Aware Minimization for Efficiently Improving Generalizationcited by 1,417
38 more
- 2016Deep CORAL: Correlation Alignment for Deep Domain Adaptationcited by 1,200
- 2020In Search of Lost Domain Generalizationcited by 1,054
- 2021Generalized Out-of-Distribution Detection: A Surveycited by 802
- 2021Generalizing to Unseen Domains: A Survey on Domain Generalizationcited by 766
- 2021Domain Generalization: A Surveycited by 764
- 2020Out-of-Distribution Generalization via Risk Extrapolation (REx)cited by 742
- 2021Domain Generalization with MixStylecited by 566
- 2019Domain Generalization by Solving Jigsaw Puzzlescited by 450
- 2017Certifying Some Distributional Robustness with Principled Adversarial Trainingcited by 337
- 2020Gradient Starvation: A Learning Proclivity in Neural Networkscited by 267
- 2021Gradient Matching for Domain Generalizationcited by 258
- 2020Deep Domain-Adversarial Image Generation for Domain Generalisationcited by 231
- 2020Domain Generalization using Causal Matchingcited by 225
- 2021Invariance Principle Meets Information Bottleneck for Out-of-Distribution Generalizationcited by 220
- 2018Composition Loss for Counting, Density Map Estimation and Localization in Dense Crowdscited by 186
- 2020Improve Unsupervised Domain Adaptation with Mixup Trainingcited by 185
- 2020Adaptive Risk Minimization: Learning to Adapt to Domain Shiftcited by 183
- 2017Domain Generalization by Marginal Transfer Learningcited by 180
- 2020Domain Adaptive Ensemble Learningcited by 156
- 2021Deep Stable Learning for Out-Of-Distribution Generalizationcited by 153
- 2021Rethinking Counting and Localization in Crowds:A Purely Point-Based Frameworkcited by 152
- 2020NWPU-Crowd: A Large-Scale Benchmark for Crowd Counting and Localizationcited by 151
- 2019Learning from Synthetic Data for Crowd Counting in the Wildcited by 148
- 2022Exact Feature Distribution Matching for Arbitrary Style Transfer and Domain Generalizationcited by 122
- 2021Open Domain Generalization with Domain-Augmented Meta-Learningcited by 115
- 2020JHU-CROWD++: Large-Scale Crowd Counting Dataset and A Benchmark Methodcited by 109
- 2020Heterogeneous Domain Generalization via Domain Mixupcited by 103
- 2020Learning to Learn with Variational Information Bottleneck for Domain Generalizationcited by 102
- 2023Sharpness-Aware Gradient Matching for Domain Generalizationcited by 102
- 2021Progressive Domain Expansion Network for Single Domain Generalizationcited by 97
- 2021Towards a Theoretical Framework of Out-of-Distribution Generalizationcited by 94
- 2018Visual Domain Adaptation with Manifold Embedded Distribution Alignmentcited by 91
- 2021Towards Principled Disentanglement for Domain Generalizationcited by 91
- 2016A Large Contextual Dataset for Classification, Detection and Counting of Cars with Deep Learningcited by 86
- 2021OoD-Bench: Quantifying and Understanding Two Dimensions of Out-of-Distribution Generalizationcited by 86
- 2021Invariant Information Bottleneck for Domain Generalizationcited by 86
- 2021Exploiting Domain-Specific Features to Enhance Domain Generalizationcited by 86
- 2021The iWildCam 2021 Competition Datasetcited by 77
50 most cited of 112 All 112 (JSON)
Built on by
Citing papers per year: 2026: 1.
1 paper.