Currently, there is an increasing demand for scientific sports training methods. Existing action analysis approaches suffer from high computational complexity, inefficient feature aggregation, and strong background interference. To address these issues, this paper proposes a spatio-temporal tube self-attention action quality assessment method for skiing sport scenarios. Firstly, a real skiing dataset containing multiple postures and scenes is constructed, annotated with Human3.6M skeleton format keypoints and expert scores. Secondly, a lightweight OpenPose improvement by using CSP-Darknet53 as the backbone is proposed, reducing computational complexity while maintaining detection accuracy. Thirdly, a Tube Self-Attention Network (TSA-Net) is designed, which generates spatio-temporal tubes via SiamMask object tracking, performs sparse self-attention feature aggregation inside the tube, and fuses pose features for quality scoring of multimodal actions. Finally, Experimental results shown that TSA-Net significantly outperforms baseline networks on skiing actions classification tasks, validating its effectiveness in optimizing computational efficiency, enhancing feature discriminability, and suppressing background interference.
Xiao-Dan Cong, Lianwu Guan, Ming Yang et al.· 2026 IEEE International Conf...· 0 citations
Accurate footprint image segmentation is challenging in forensic applications because fine anatomical structures, weak boundaries, and background interference can degrade segmentation performance. This study presents a task-oriented generative adversarial network (GAN)-based framework for forensic footprint image segmentation. Channel Prior Convolutional Attention (CPCA) modules are integrated into the decoder stages of the generator to recalibrate fused encoder–decoder features and preserve fine details in the toe, arch, and heel regions. In addition, a dual-branch discriminator processes image–mask pairs at the original and downsampled scales, providing complementary constraints on local boundary details and global footprint morphology. The framework is trained with a least-squares adversarial loss and a binary cross-entropy (BCE)–Dice segmentation loss. Experiments on the self-collected aFoot_2025 dataset show that the proposed framework achieves an IoU of 0.9448 and a Dice coefficient of 0.9713, outperforming the evaluated baseline and attention-based alternatives. Under the evaluated synthetic Gaussian-noise settings, the proposed method retained relatively stable segmentation performance. Furthermore, an exploratory footprint-based height-prediction analysis showed modestly lower prediction errors than the baseline GAN. These findings indicate that, under the controlled acquisition conditions of the aFoot_2025 dataset, CPCA-based feature calibration and dual-scale discrimination may improve segmentation-mask quality and provide a possible benefit for subsequent anthropometric analysis.
