FDC-pose: Enhancing multi-person pose estimation in complex scenes via dynamic synergy mechanisms

Mengye Lin; Zhiming Cai; Yixin Zhang; Jiangchao Zhang; Zukun Xu; Huabin He; Mengye Lin; Zhiming Cai; Yixin Zhang; Jiangchao Zhang; Zukun Xu; Huabin He

doi:10.3934/era.2026260

Electronic Research Archive

2026, Volume 34, Issue 9: 5861-5886. doi: 10.3934/era.2026260

Previous Article Next Article

Research article Special Issues

FDC-pose: Enhancing multi-person pose estimation in complex scenes via dynamic synergy mechanisms

1.
School of Electrical and Information Engineering, Fujian University of Technology, Fuzhou 350118, China
2.
National Demonstration Center for Experimental Electronic Information and Electrical Technology Education, Fujian University of Technology, Fuzhou 350118, China

Received: 25 April 2026 Revised: 26 June 2026 Accepted: 30 June 2026 Published: 09 July 2026

Real-time multi-person pose estimation suffers from inter-instance occlusion and severe scale variations. To resolve these bottlenecks, we present focal-dynamic context (FDC)-pose, an efficient single-stage architecture tailored for crowded scenes. The framework incorporates three sequential innovations: 1) a cascaded pyramid network-focal omni-dimensional dynamic convolution (CPN-FocalOD) backbone combining context modulation with omni-dimensional dynamic filtering to disentangle overlapping silhouettes; 2) a cross-scale fusion network featuring content-aware upsampling (DySample) to suppress quantization errors and preserve distal joint details; and 3) a minimum point distance intersection over union (MPDIoU) regression loss enforcing point-level geometric constraints to eliminate occlusion-induced localization jitter. Evaluations on the common objects in context (COCO) 2017 dataset demonstrate that FDC-pose yields absolute gains of 3.4% in AP, 3.0% in AP₅₀, and 4.2% in the strict AP$ _{75} $ metric relative to the baseline. Parallel evaluations on the CrowdPose dataset show that our framework presents a prominent robustness gain, yielding an absolute improvement of 4.5% in AP.
- human pose estimation,
- YOLO,
- dynamic convolution,
- multi-scale feature fusion,
- content-aware upsampling,
- bounding box regression
Citation: Mengye Lin, Zhiming Cai, Yixin Zhang, Jiangchao Zhang, Zukun Xu, Huabin He. FDC-pose: Enhancing multi-person pose estimation in complex scenes via dynamic synergy mechanisms[J]. Electronic Research Archive, 2026, 34(9): 5861-5886. doi: 10.3934/era.2026260

Related Papers:

Abstract

Real-time multi-person pose estimation suffers from inter-instance occlusion and severe scale variations. To resolve these bottlenecks, we present focal-dynamic context (FDC)-pose, an efficient single-stage architecture tailored for crowded scenes. The framework incorporates three sequential innovations: 1) a cascaded pyramid network-focal omni-dimensional dynamic convolution (CPN-FocalOD) backbone combining context modulation with omni-dimensional dynamic filtering to disentangle overlapping silhouettes; 2) a cross-scale fusion network featuring content-aware upsampling (DySample) to suppress quantization errors and preserve distal joint details; and 3) a minimum point distance intersection over union (MPDIoU) regression loss enforcing point-level geometric constraints to eliminate occlusion-induced localization jitter. Evaluations on the common objects in context (COCO) 2017 dataset demonstrate that FDC-pose yields absolute gains of 3.4% in AP, 3.0% in AP₅₀, and 4.2% in the strict AP$ _{75} $ metric relative to the baseline. Parallel evaluations on the CrowdPose dataset show that our framework presents a prominent robustness gain, yielding an absolute improvement of 4.5% in AP.

References

[1]	S. Ren, K. He, R. Girshick, J. Sun, Faster R-CNN: Towards real-time object detection with region proposal networks, IEEE Trans. Pattern Anal. Mach. Intell., 39 (2017), 1137–1149. https://doi.org/10.1109/TPAMI.2016.2577031 doi: 10.1109/TPAMI.2016.2577031
[2]	C. Zheng, W. Wu, C. Chen, T. Yang, S. Zhu, J. Shen, et al., Deep learning-based human pose estimation: A survey, ACM Comput. Surv., 56 (2023), 1–37. https://doi.org/10.1145/3603618 doi: 10.1145/3603618
[3]	M. Ma, X. He, X. Bai, A survey of deep learning-based human pose estimation: Methods, datasets, and evaluation metrics, in 2025 2nd International Conference on Digital Image Processing and Computer Applications (DIPCA), (2025), 146–152. https://doi.org/10.1109/DIPCA65051.2025.11042554
[4]	Y. Chen, Z. Zhang, C. Yuan, B. Li, Y. Deng, W. Hu, Channel-wise topology refinement graph convolution for skeleton-based action recognition, preprint, arXiv:2107.12213.
[5]	Y. Wang, P. Liu, H. Kang, D. Wu, D. Miao, ICFNet: Interactive-complementary fusion network for monocular 3D human pose estimation, Neurocomputing, 616 (2025), 128947. https://doi.org/10.1016/j.neucom.2024.128947 doi: 10.1016/j.neucom.2024.128947
[6]	A. Newell, K. Yang, J. Deng, Stacked hourglass networks for human pose estimation, in Computer Vision-ECCV 2016, 9912 (2016), 483–499. https://doi.org/10.1007/978-3-319-46484-8_29
[7]	B. Xiao, H. Wu, Y. Wei, Simple baselines for human pose estimation and tracking, in Computer Vision-ECCV 2018, 11210 (2018), 472–487. https://doi.org/10.1007/978-3-030-01231-1_29
[8]	Y. Chen, Z. Wang, Y. Peng, Z. Zhang, G. Yu, J. Sun, Cascaded pyramid network for multi-person pose estimation, in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2018), 7103–7112. https://doi.org/10.1109/CVPR.2018.00742
[9]	K. Sun, B. Xiao, D. Liu, J. Wang, Deep high-resolution representation learning for human pose estimation, in 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2019), 5686–5696. https://doi.org/10.1109/CVPR.2019.00584
[10]	Y. Xu, J. Zhang, Q. Zhang, D. Tao, ViTPose: Simple vision transformer baselines for human pose estimation, preprint, arXiv:2204.12484.
[11]	D. Yang, Y. Ge, N. Xu, R. Shi, DDCEFormer: Dual-domain cross enhanced transformer for 3D human pose estimation, Neurocomputing, 681 (2026), 133333. https://doi.org/10.1016/j.neucom.2026.133333 doi: 10.1016/j.neucom.2026.133333
[12]	C. Yu, B. Xiao, C. Gao, L. Yuan, L. Zhang, N. Sang, et al., Lite-HRNet: A lightweight high-resolution network, in 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2021), 10435–10445. https://doi.org/10.1109/CVPR46437.2021.01030
[13]	Y. Chen, Y. Tian, M. He, Monocular human pose estimation: A survey of deep learning-based methods, Comput. Vision Image Understanding, 192 (2020), 102897. https://doi.org/10.1016/j.cviu.2019.102897 doi: 10.1016/j.cviu.2019.102897
[14]	Z. Geng, K. Sun, B. Xiao, Z. Zhang, J. Wang, Bottom-up human pose estimation via disentangled keypoint regression, preprint, arXiv:2104.02300.
[15]	Z. Cao, G. Hidalgo, T. Simon, S. Wei, Y. Sheikh, OpenPose: Realtime multi-person 2D pose estimation using part affinity fields, IEEE Trans. Pattern Anal. Mach. Intell., 43 (2021), 172–186. https://doi.org/10.1109/TPAMI.2019.2929257 doi: 10.1109/TPAMI.2019.2929257
[16]	C. Wang, A. Bochkovskiy, H. M. Liao, YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors, in 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2023), 7464–7475. https://doi.org/10.1109/CVPR52729.2023.00721
[17]	T. Jiang, P. Lu, L. Zhang, N. Ma, R. Han, C. Lyu, et al., RTMPose: Real-time multi-person pose estimation based on MMPose, preprint, arXiv:2303.07399.
[18]	Y. Chen, X. Dai, M. Liu, D. Chen, L. Yuan, Z. Liu, Dynamic convolution: Attention over convolution kernels, in 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2020), 11027–11036. https://doi.org/10.1109/CVPR42600.2020.01104
[19]	C. Li, A. Zhou, A. Yao, Omni-dimensional dynamic convolution, preprint, arXiv:2209.07947.
[20]	J. Wang, K. Chen, R. Xu, Z. Liu, C. C. Loy, D. Lin, CARAFE: Content-aware reassembly of features, preprint, arXiv:1905.02188.
[21]	S. Liu, D. Huang, Y. Wang, Learning spatial fusion for single-shot object detection, preprint, arXiv:1911.09516.
[22]	Z. Zheng, P. Wang, D. Ren, W. Liu, R. Ye, Q. Hu, et al., Enhancing geometric factors in model learning and inference for object detection and instance segmentation, IEEE Trans. Cybern., 52 (2022), 8574–8586. https://doi.org/10.1109/TCYB.2021.3095305 doi: 10.1109/TCYB.2021.3095305
[23]	G. Lan, Y. Wu, F. Hu, Q. Hao, Vision-based human pose estimation via deep learning: A survey, IEEE Trans. Hum.-Mach. Syst., 53 (2023), 253–268. https://doi.org/10.1109/THMS.2022.3219242 doi: 10.1109/THMS.2022.3219242
[24]	D. Maji, S. Nagori, M. Mathew, D. Poddar, YOLO-Pose: Enhancing YOLO for multi person pose estimation using object keypoint similarity loss, preprint, arXiv:2204.06806.
[25]	J. Chen, X. Wang, Z. Guo, X. Zhang, J. Sun, Dynamic region-aware convolution, in 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2021), 8060–8069. https://doi.org/10.1109/CVPR46437.2021.00797
[26]	J. Yang, C. Li, X. Dai, L. Yuan, J. Gao, Focal modulation networks, preprint, arXiv:2203.11926.
[27]	S. Liu, L. Qi, H. Qin, J. Shi, J. Jia, Path aggregation network for instance segmentation, in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2018), 8759–8768. https://doi.org/10.1109/CVPR.2018.00913
[28]	S. Huang, Z. Lu, R. Cheng, C. He, FaPN: Feature-aligned pyramid network for dense image prediction, in 2021 IEEE/CVF International Conference on Computer Vision (ICCV), (2021), 844–853. https://doi.org/10.1109/ICCV48922.2021.00090
[29]	W. Liu, H. Lu, H. Fu, Z. Cao, Learning to upsample by learning to sample, in 2023 IEEE/CVF International Conference on Computer Vision (ICCV), (2023), 6004–6014. https://doi.org/10.1109/ICCV51070.2023.00554
[30]	H. Rezatofighi, N. Tsoi, J. Gwak, A. Sadeghian, I. Reid, S. Savarese, Generalized intersection over union: A metric and a loss for bounding box regression, in 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2019), 658–666. https://doi.org/10.1109/CVPR.2019.00075
[31]	X. Li, W. Wang, L. Wu, S. Chen, X. Hu, J. Li, et al., Generalized focal loss: Learning qualified and distributed bounding boxes for dense object detection, preprint, arXiv:2006.04388.
[32]	S. Ma, Y. Xu, MPDIoU: A loss for efficient and accurate bounding box regression, preprint, arXiv:2307.07662.
[33]	J. Li, S. Bian, A. Zeng, C. Wang, B. Pang, W. Liu, et al., Human pose regression with residual log-likelihood estimation, preprint, arXiv:2107.11291.
[34]	Y. Ma, L. Xu, Y. Zhang, T. Zhang, X. Luo, Steganalysis feature selection with multidimensional evaluation and dynamic threshold allocation, IEEE Trans. Circuits Syst. Video Technol., 34 (2024), 1954–1969. https://doi.org/10.1109/TCSVT.2023.3295364 doi: 10.1109/TCSVT.2023.3295364
[35]	Z. Liu, H. Mao, C. Wu, C. Feichtenhofer, T. Darrell, S. Xie, A ConvNet for the 2020s, in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2022), 11966–11976. https://doi.org/10.1109/CVPR52688.2022.01167
[36]	G. Huang, Y. Sun, Z. Liu, D. Sedra, K. Q. Weinberger, Deep networks with stochastic depth, in Computer Vision-ECCV 2016, 9908 (2016), 646–661. https://doi.org/10.1007/978-3-319-46493-0_39
[37]	Y. Ma, L. Xu, Q. Zhang, Y. Zhang, X. Xin, X. Luo, EIS-OBEA: Enhanced image steganalysis via opposition-based evolutionary algorithm, IEEE Trans. Inf. Forensics Secur., 20 (2025), 3616–3631. https://doi.org/10.1109/TIFS.2025.3549692 doi: 10.1109/TIFS.2025.3549692
[38]	Y. Zhao, W. Lv, S. Xu, J. Wei, G. Wang, Q. Dang, et al., DETRs beat YOLOs on real-time object detection, preprint, arXiv:2304.08069.
[39]	R. Khanam, M. Hussain, YOLOv11: An overview of the key architectural enhancements, preprint, arXiv:2410.17725.
[40]	H. Fang, J. Li, H. Tang, C. Xu, H. Zhu, Y. Xiu, et al., AlphaPose: Whole-body regional multi-person pose estimation and tracking in real-time, IEEE Trans. Pattern Anal. Mach. Intell., 45 (2023), 7157–7173. https://doi.org/10.1109/TPAMI.2022.3222784 doi: 10.1109/TPAMI.2022.3222784
[41]	T. Lin, M. Maire, S. Belongie, L. Bourdev, R. Girshick, J. Hays, et al., Microsoft COCO: Common objects in context, preprint, arXiv:1405.0312.
[42]	J. Li, C. Wang, H. Zhu, Y. Mao, H. Fang, C. Lu, CrowdPose: Efficient crowded scenes pose estimation and a new benchmark, preprint, arXiv:1812.00324.
[43]	F. Yu, D. Wang, E. Shelhamer, T. Darrell, Deep layer aggregation, in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2018), 2403–2412. https://doi.org/10.1109/CVPR.2018.00255
[44]	C. Neff, A. Sheth, S. Furgurson, J. Middleton, H. Tabkhi, EfficientHRNet: Efficient and scalable high-resolution networks for real-time multi-person 2D human pose estimation, J. Real-Time Image Process., 18 (2021), 1037–1049. https://doi.org/10.1007/s11554-021-01132-9 doi: 10.1007/s11554-021-01132-9
[45]	H. Wang, J. Liu, J. Tang, G. Wu, Lightweight super-resolution head for human pose estimation, in Proceedings of the 31st ACM International Conference on Multimedia, (2023), 2353–2361. https://doi.org/10.1145/3581783.3612236
[46]	J. Terven, D. Córdova-Esparza, J. Romero-González, A comprehensive review of YOLO architectures in computer vision: From YOLOv1 to YOLOv8 and YOLO-NAS, Mach. Learn. Knowl. Extr., 5 (2023), 1680–1716. https://doi.org/10.3390/make5040083 doi: 10.3390/make5040083
[47]	Y. Tang, K. Han, J. Guo, C. Xu, C. Xu, Y. Wang, GhostNetV2: Enhance cheap operation with long-range attention, preprint, arXiv:2211.12905.
[48]	S. Woo, S. Debnath, R. Hu, X. Chen, Z. Liu, I. S. Kweon, et al., ConvNeXt V2: Co-designing and scaling ConvNets with masked autoencoders, in 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2023), 16133–16142. https://doi.org/10.1109/CVPR52729.2023.01548
[49]	S. Mehta, M. Rastegari, MobileViT: Light-weight, general-purpose, and mobile-friendly vision transformer, preprint, arXiv:2110.02178.
[50]	M. Maaz, A. Shaker, H. Cholakkal, S. Khan, S. W. Zamir, R. M. Anwer, et al., EdgeNeXt: Efficiently amalgamated CNN-transformer architecture for mobile vision applications, preprint, arXiv:2206.10589.
[51]	X. Ding, X. Zhang, J. Han, G. Ding, Scaling up your kernels to $31\times31$: Revisiting large kernel design in CNNs, in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2022), 11953–11965. https://doi.org/10.1109/CVPR52688.2022.01166
[52]	J. He, S. Erfani, X. Ma, J. Bailey, Y. Chi, X. Hua, Alpha-IoU: A family of power intersection over union losses for bounding box regression, preprint, arXiv:2110.13675.
[53]	H. Zhang, S. Zhang, Shape-IoU: More accurate metric considering bounding box shape and scale, preprint, arXiv:2312.17663.
[54]	Z. Gevorgyan, SIoU loss: More powerful learning for bounding box regression, preprint, arXiv:2205.12740.
[55]	Y. Zhang, W. Ren, Z. Zhang, Z. Jia, L. Wang, T. Tan, Focal and efficient IOU loss for accurate bounding box regression, preprint, arXiv:2101.08158.
[56]	Z. Tong, Y. Chen, Z. Xu, R. Yu, Wise-IoU: Bounding box regression loss with dynamic focusing mechanism, preprint, arXiv:2301.10051.

Reader Comments

Your name:*

Email:*
© 2026 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)