Chengjian Li

Text-driven diffusion models have achieved remarkable performance in human motion generation. However, these generative works struggle to generate high-quality motion consistent with textual descriptions. The primary reasons are: 1) insufficient fine-grained motion modeling due to the motion representations being difficult to distinguish in latent diffusion; and 2) inconsistencies between motions and textual descriptions due to misalignment in the multimodal space. To overcome these limitations, this work proposes the Motion generation with Frequency and Text State Space models (MoFTSS) including two main modules: frequency state space model (FreqSSM) and text state space model (TextSSM). Specifically, FreqSSM derives fine-grained representations by decomposing sequences into low-frequency and high-frequency components. This allows it to guide the generation of static poses (e.g., sitting, lying) and fine-grained motions (e.g, transitions, stumbling). For consistency between text and motion, TextSSM treats text features as a semantic modulation term within the SSM, enabling dynamic filtering of motion features consistent with textual semantics. Extensive experiments suggest that our MoFTSS achieves superior performance on the text-to-motion generation task. Notably, it attains the lowest FID of 0.181 on the HumanML3D dataset, significantly lower than the 0.421 achieved by MLD.

As the two most typical classic network models, Convolutional Neural Networks (CNN) and Transformer have been widely applied in Obstructive sleep apnea (OSA) detection in recent years. However, due to the inherent limitations of the receptive field in traditional CNN models (the receptive field is positively correlated with the fixed convolution kernel size, and the ability to extract global feature information is limited), further improvement in their performance is constrained. While, for Transformer, due to the computational complexity of the self attention mechanism in the Transformer model increases exponentially with the length of the context, it will hold a very high computational overhead, and which would hinder the deployment of transformer on devices with limited computing resources. To address these problems, this paper proposes an adaptive bidirectional selective state space model based method for OSA detection, termed as OSAMamba. The main novelty of the proposed method lies in the following two aspects: the development of the lightweight multi-scale efficient aggregation (LMSEA) module, and the propose of the adaptive bidirectional selective state space model (ABSM). To achieve the purpose of expanding the model receptive field and capturing the effective temporal features with a very low number of parameters, the LMSEA module adopts a combination of partial convolution (PConv)-based multiscale strategy and convolutional block attention module (CBAM). The purpose of the ABSM module is to reduce the computational cost of the model and improve the model deployability by using a frequency domain enhancement strategy to fuse the effective time domain features extracted by adaptive bi-directional Mamba (ABi-Mamba) with linear complexity with the frequency domain features extracted by the Frequency Domain Enhancement Module (FEM). Extensive experiments on the Apnea-ECG dataset show that of all compared methods, the proposed method obtains the best accuracy of 91.91% in the per-segment detection, and which surpassing the state-of-the-art (SOTA) TFFormer by 0.31%. It also achieves a remarkable accuracy of 100% with the lowest mean absolute error (MAE) of 2.43 in per-record detection.

Although Transformer-based methods have achieved significant success in Obstructive sleep apnea (OSA) detection, they suffer from higher computational costs, lower feature capturing ability in frequency domain, and class imbalance at the data level. To address these problems, this paper proposes a data rebalancing loss constrained light transformer for OSA detection, called DRLFormer. The core of the proposed method lies in the following two aspects: the development of the Additive Temporal-Frequency Fusion Attention (ATFA) module, and the propose of Data Re-balancing Loss (DRLoss) function. The purpose of the ATFA module is to fuse the frequency domain features extracted by the frequency domain enhancement module (FEM) with the important time domain features extracted by the learnable time domain attention (LTA) in a very low computational cost by novel using the summation strategy. The purpose of DRLoss is to solve the problem of imbalanced samples within batches which usually ignored by existing common used CrossEntropy loss (CELoss) . the novelty of the designing of DRLoss lies in by designing weight memory units (WMU), and allocatting different weights for correctly and incorrectly predicted samples to achieve prediction level balance through classification penalty item (CPI). Extensive experiments are conducted on the Apnea-ECG dataset, and compared with the state-of-the-art method. Our method achieving an improving by 0.44% in accuracy (from 91.68% improved to 92.12%) with a reduction of parameter size by 74% and FLOPs by 53%. Furthermore, the excellent performance was achieved on UCD, PhysioNet, and clinical XJ300 datasets, thoroughly verifying the effectiveness and clinical application potential of DRLFormer.

Accurate detection of Obstructive Sleep Apnea (OSA) with a single-lead electrocardiogram (ECG) signal is highly desirable for timely treating of OSA patients. However, due to the variance of apneas in appearance and size in ECG signals, it is still a very challenging task to obtain an accurate OSA apnea detection. To address this problem, this paper presents a time frequency information fusion based CNN-Transformer model (TFFormer) for OSA detection with Single-lead ECG. In which, a module consisting of a deep residual shrinkage module, a multi-scale convolutional attention module (MSCA), and a multi-layer convolution module is developed for time-frequency feature extraction. The purpose of this operation is to extract rich features from a short length of ECG signal sequences with a low computation cost. For time-frequency information fusion, to reduce its computation cost, a gated self-attention based adaptive pruning time-frequency information fusion module is developed to prune the redundant tokens. With the attention based adaptive pruning time-frequency information fusion module(APTFFA), the TFFormer is constructed for data parallel processing and long-distance modeling. Compared with the best model in the comparative method, the accuracy of the proposed method was improved by 0.18% in the segmented case, and the mean absolute error was reduced by 0.25 per-recorded case, which demonstrates that the TFFormer model has better OSA detection performance and could provide a convenient and accurate solution for clinical OSA detection.