Abstract
Heavily occluded pedestrian detection remains challenging for CNN detectors. Recent methods such as OTA and simOTA utilize optimal transport for label assignment but still encounter limitations in handling local occlusion. To tackle this issue, we thoroughly investigate the relationship between data assignment algorithms and the label assignment problem. We propose a theoretical framework to explain the underlying causes of suboptimal label assignments in heavily occluded regions and identify the ideal assignment method. In our pursuit of the ideal method, we propose two label assignment methods: the K-means method (KMM) and the LAPJV method (LAM), which correspond to the Clustering Algorithm and the Linear Assignment Problem, respectively. KMM assigns anchors based on the lowest cost, similar to K-means clustering. LAM applies LAPJV iteratively on occluded regions for local optimization, and maintains global optimality in non-occluded regions. LAM also achieves 30% execution time reduction compared to OTA. We provide both theoretical analysis and experimental validation to demonstrate that LAM is the ideal method in our theoretical framework. It elegantly reconciles global and local optimal assignments efficiently, thus achieving the highest performance in Average Precision (AP) and Recall on five datasets, i.e., CrowdHuman, WiderPerson, CityPersons, COCOPersons, and COCO.
Similar content being viewed by others
Data availability
No datasets were generated or analyzed during the current study.
Notes
In general, different outputs of a detector are responsible for detecting objects of different scales. Therefore, assigning ground truths with similar scales to anchors from the same output is crucial for effective training.
COCOPersons is a subset of COCO, where only annotations of “person” are considered for training and evaluation.
The test set does not provide annotations and the server is no longer accessible.
References
He, Y., He, N., Yu, H., Zhang, R., Yan, K.: From macro to micro: rethinking multi-scale pedestrian detection. Multimed. Syst. 29(3), 1417–1429 (2023)
Chu, X., Zheng, A., Zhang, X., Sun, J.: Detection in crowded scenes: one proposal, multiple predictions. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 12214–12223 (2020)
Liu, S., Huang, D., Wang, Y.: Adaptive NMS: refining pedestrian detection in a crowd. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 6459–6468 (2019)
Zhang, S., Yang, J., Schiele, B.: Occluded pedestrian detection through guided attention in CNNS. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 6995–7003 (2018)
Zhou, C., Yuan, J.: Bi-box regression for pedestrian detection and occlusion estimation. In: Proceedings of the European Conference on Computer Vision (ECCV), pp. 135–151 (2018)
Ge, Z., Liu, S., Li, Z., Yoshie, O., Sun, J.: Ota: Optimal transport assignment for object detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 303–312 (2021)
Ge, Z., Liu, S., Wang, F., Li, Z., Sun, J.: Yolox: exceeding yolo series in 2021. arXiv preprint arXiv:2107.08430 (2021)
Jain, A.K., Murty, M.N., Flynn, P.J.: Data clustering: a review. ACM Comput. Surv. 31(3), 264–323 (1999). https://doi.org/10.1145/331499.331504
Martello, S., Toth, P.: Linear assignment problems. In: Martello, S., Laporte, G., Minoux, M., Ribeiro, C. (eds.) Surveys in Combinatorial Optimization. North-Holland Mathematics Studies, vol. 132, pp. 259–282. North-Holland, Amsterdam (1987)
Peyré, G., Cuturi, M., et al.: Computational optimal transport: With applications to data science. Foundations Trends Mach. Learn. 11(5–6), 355–607 (2019)
Jonker, R., Volgenant, T.: A shortest augmenting path algorithm for dense and sparse linear assignment problems. In: DGOR/NSOR: Papers of the 16th Annual Meeting of DGOR in Cooperation with NSOR/Vorträge der 16. Jahrestagung der DGOR Zusammen Mit der NSOR, p. 622. Springer (1988)
Cuturi, M.: Sinkhorn distances: lightspeed computation of optimal transport. Adv. Neural Inf. Process. Syst. 26, 2292–2300 (2013)
Tian, Z., Shen, C., Chen, H., He, T.: Fcos: fully convolutional one-stage object detection. In: Proceedings of the IEEE International Conference on Computer Vision, pp. 9627–9636 (2019)
Zhang, S., Chi, C., Yao, Y., Lei, Z., Li, S.Z.: Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 9759–9768 (2020)
Shao, S., Zhao, Z., Li, B., Xiao, T., Yu, G., Zhang, X., Sun, J.: Crowdhuman: A benchmark for detecting human in a crowd. arXiv preprint arXiv:1805.00123 (2018)
Zhang, S., Xie, Y., Wan, J., Xia, H., Li, S.Z., Guo, G.: Widerperson: a diverse dataset for dense pedestrian detection in the wild. IEEE Trans. Multimed. 22(2), 380–393 (2019)
Zhang, S., Benenson, R., Schiele, B.: Citypersons: a diverse dataset for pedestrian detection. In: CVPR, pp. 3213–3221 (2017)
Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., Zitnick, C.L.: Microsoft coco: common objects in context. In: European Conference on Computer Vision, pp. 740–755. Springer (2014)
Gao, Z., Chen, P., Zhuo, T., Liu, M., Zhu, L., Wang, M., Chen, S.: A semantic perception and cnn-transformer hybrid network for occluded person re-identification. IEEE Trans. Circuits Syst. Video Technol. (2023). https://doi.org/10.1109/TCSVT.2023.3296680
Zhang, S., Wen, L., Bian, X., Lei, Z., Li, S.Z.: Occlusion-aware r-cnn: detecting pedestrians in a crowd. In: Proceedings of the European Conference on Computer Vision (ECCV), pp. 637–653 (2018)
He, Y., Zhu, C., Yin, X.-C.: Occluded pedestrian detection via distribution-based mutual-supervised feature learning. IEEE Trans. Intell. Transp. Syst. 23(8), 10514–10529 (2022). https://doi.org/10.1109/TITS.2021.3094800
Kuhn, H.W.: The Hungarian method for the assignment problem. Naval Res. Logist. Quart. 2(1–2), 83–97 (1955)
Bertsekas, D.P.: Auction algorithms for network flow problems: a tutorial introduction. Comput. Optim. Appl. 1(1), 7–66 (1992)
Frogner, C., Zhang, C., Mobahi, H., Araya, M., Poggio, T.A.: Learning with a wasserstein loss. In: Cortes, C., Lawrence, N., Lee, D., Sugiyama, M., Garnett, R. (eds.) Advances in Neural Information Processing Systems, vol. 28 (2015). https://proceedings.neurips.cc/paper/2015/file/a9eb812238f753132652ae09963a05e9-Paper.pdf
Benamou, J.-D., Carlier, G., Cuturi, M., Nenna, L., Peyré, G.: Iterative Bregman projections for regularized transportation problems. SIAM J. Sci. Comput. 37(2), 1111–1138 (2015)
He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: CVPR, pp. 770–778 (2016)
Zhang, X., Wan, F., Liu, C., Ji, R., Ye, Q.: Freeanchor: learning to match anchors for visual object detection. In: Advances in Neural Information Processing Systems, pp. 147–155 (2019)
Kim, K., Lee, H.S.: Probabilistic anchor assignment with IOU prediction for object detection. In: Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXV 16, pp. 355–371. Springer (2020)
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., Fei-Fei, L.: Imagenet: a large-scale hierarchical image database. In: 2009 IEEE Conference on Computer Vision and Pattern Recognition, pp. 248–255. IEEE (2009)
Lin, T.-Y., Dollár, P., Girshick, R., He, K., Hariharan, B., Belongie, S.: Feature pyramid networks for object detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2117–2125 (2017)
Acknowledgements
This work is supported in part by the National Natural Science Foundation of China (NSFC) under Grant (No.61932020), and the Taishan Scholar Program of Shandong Province (tstp20221128).
Author information
Authors and Affiliations
Contributions
C.L. wrote the main manuscript text and prepared figures and tables. All the authors reviewed the manuscript.
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no conflict of interest.
Additional information
Communicated by T. Li.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Liu, C., Li, H., Wang, Z. et al. Reconciling global and local optimal label assignments for heavily occluded pedestrian detection. Multimedia Systems 30, 100 (2024). https://doi.org/10.1007/s00530-024-01304-0
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00530-024-01304-0