SMSongping Mai

Songping Mai

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Songping Mai has a total of 76 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • High-precision and efficient single-shot color fringe projection profilometry via implicit spectral extraction and fusion network

    by Yongliang Ye, Qiming Zhu, Junwei Hu, Songping Mai

    Single-shot color fringe projection profilometry (SCFPP) is extensively researched for rapid 3D reconstruction but faces accuracy challenges on discontinuous surfaces and in multi-channel color projection due to crosstalk. To address these issues, this paper proposes a high-precision and efficient implicit spectral extraction (ISE) and fusion network. Specifically, an ISE block is introduced to overcome the inherent limitations of transform-based methods on discontinuous surfaces. By approximating the frequency-domain transformation using only basic neural units, the ISE block achieves implicit spectral representation without complex explicit transforms. A multi-path feature fusion (MPF) block, integrating dual attention paths and a residual path, is further designed to enhance feature fusion and color crosstalk suppression. In our SCFPP system, the network directly maps a single color composite fringe pattern to three groups of phase sine and cosine maps, from which the wrapped phase (via arctangent function) and absolute phase (via three-frequency heterodyne unwrapping) are sequentially derived for 3D reconstruction. For training and evaluation, a dataset with varying intrinsic and extrinsic parameters was collected. Experimental results demonstrate that the proposed method achieves higher precision in single-shot phase decoding and 3D reconstruction while maintaining computational efficiency, attaining a planar distance measurement error of only 0.0939 mm, which represents the best result among all evaluated single-shot methods.

  • Open Access

    GhostIC: A Ghost-Driven Lightweight Network for Efficient Learned Image Compression

    by Jie Yu, Songping Mai, Peng Zhang, Xiaofeng Huang, Heming Sun, Jian Cheng, Yucheng Jiang

    Learned image compression (LIC) models have achieved remarkable performance but typically rely on computationally expensive backbones, which limits their applicability on resource-constrained embedded devices. To address this limitation, we present GhostIC, a Ghost-driven lightweight image compression network that achieves efficient inference while maintaining competitive rate–distortion performance. In GhostIC, we design a GhostBlock inspired by GhostNet where redundant features are generated from intrinsic features using inexpensive operations. This design integrates prior knowledge of image spatial redundancy, enabling efficient representation for image compression. Furthermore, the GhostBlock integrates the decoupled fully connected (DFC) attention mechanism, which enhances long-range dependency modeling with minimal overhead and compensates for the representational limitations of lightweight convolutions. Extensive experiments on standard benchmarks demonstrate that GhostIC delivers a rate-distortion performance on par with VVC intra coding while substantially reducing computational complexity compared to state-of-the-art learned models. With its favorable balance between efficiency and performance, GhostIC provides a practical solution for real-time learned image compression on resource-constrained devices. The source code will be made available upon publication.

  • CoCRNet: color crosstalk removal network for color fringe projection profilometry

    by Qiming Zhu, Yongliang Ye, Junwei Hu, Songping Mai

    Color crosstalk severely limits the practical accuracy of color fringe projection profilometry (color FPP). Existing compensation methods typically rely on calibration procedures or simplified analytic models, while recent learning-based approaches tend to directly regress phase from color fringe images, without explicitly addressing the underlying color crosstalk in the fringe domain. In this paper, we formulate color crosstalk suppression as an image-level fringe restoration task, aiming to recover crosstalk-suppressed color fringe images from real captured observations while remaining fully compatible with conventional FPP pipelines. Building upon a U-Net backbone, we propose CoCRNet with a series of task-specific modular designs, including exposure-aware feature normalization to alleviate brightness variations and explicit color-channel interaction modeling to suppress inter-channel interference in the fringe domain. The proposed network is trained and evaluated entirely on real scenes, with additional multi-level exposure simulation applied during training to further enhance robustness. Experimental results on real datasets demonstrate that CoCRNet outperforms existing learning-based methods in both fringe domain color crosstalk suppression and downstream phase reconstruction, achieving lower wrapped phase, fringe order, and unwrapped phase errors.

  • A review of wireless power and data transfer systems for biomedical implants

    by Binyao Hong, Xiao Wei, Chengyu Shi, Songpin Mai, Xian Tang, Zhihua Wang

    This paper reviews recent research on wireless power and data transfer (WPDT) systems for implanted medical devices (IMDs). Focusing primarily on inductive WPDT systems, the review incorporates theoretical analyses and discussion of link optimization strategies. These strategies target power transfer efficiency (PTE) degradation caused by impedance mismatch, coil misalignment, inter-coil distance, and coupling angle-induced magnetic field inhomogeneity, summarizing referable solutions to mitigate such performance losses. The review also details key WPDT system components: in the power path, it covers power amplifiers, rectifiers, and voltage regulators; in the data path, it involves modulation schemes such as Amplitude-Shift Keying (ASK), Phase-Shift Keying (PSK), Frequency-Shift Keying (FSK), and Load-Shift Keying (LSK). Addressing the core challenge of balancing high PTE (typ. 50%) and data rate (typ. 0.1–2Mbps) under dynamic coupling and load variations, it summarizes the circuit innovation directions of each component, extending to integrated innovation paths at the system level. Finally, future directions are outlined, focusing on miniaturization, efficiency optimization via advanced circuits, biosafety, and robust modulation to enhance data reliability and speed, as well as the deep integration of machine learning for performance improvement.


  • PAA-RLME: A Robust Online Adaptive Algorithm for Nonlinear Error Suppression in Phase-Shifting Profilometry

    by Jw Hu, Yucheng Jiang, Songping Mai, Yu Yu, Yongliang Ye

    This study proposes an online blind estimation method for nonlinear distortion parameters in structured light measurement systems, addressing key limitations in three-step phase-shifting profilometry (PSP). The method overcomes the challenges of nonlinear intensity response from projection devices, which affects measurement accuracy. Current solutions, such as traditional pre-calibration and blind nonlinear compensation approaches, either require cumbersome procedures or have limited adaptability and accuracy due to dependence on the surface profile of the measured object. The proposed PAA-RLME (Polar Angle Average - Robust LM Parameter Estimation) algorithm uses polar angle average (PAA) and robust $3\sigma$ screening to estimate nonlinear parameters, eliminating the need for time-consuming calibration. It integrates parameter estimation with measurement, enhancing accuracy, simplifying device design, and maintaining efficiency. Compared to existing online nonlinear error correction methods, the approach is based on parameter estimation, improving both compensation accuracy and robustness. Validation through theoretical analysis, simulations, and experiments shows the method’s effectiveness. It reduces absolute mean phase error by up to 57\% under real-world conditions, with compensation accuracy decoupled from the object’s surface, demonstrating improved stability over previous methods.

  • A Dual-mode FSK Wireless Power and Data Transfer System with 1.1 Mbps Data Rate and Rectifier Output Power of 148 mW for Implantable Medical Applications

    by Congyi Qian, Sai Li, Heng Liang, Hongyuan Zhang, Songping Mai

    This paper proposes a wireless power and data transfer (WPDT) system for implantable medical applications, featuring a simple structure, high data rate (DR), and efficient power transmission. To streamline the frequency-shift keying (FSK) data transmission link, the FSK modulator integrates merely one oscillator and one frequency divider, while the FSK demodulator requires only one D flip-flop and one delay unit. This minimalist design generates two FSK carrier signals with a large frequency difference, simultaneously enhancing the data transmission rate and reducing the bit error rate (BER). To meet the WPDT system's requirements for high power transmission under transient conditions and low coupling coefficients, a coupled network capacitive compensation technique is employed. This method significantly enhances the power transmission capability of one carrier frequency, enabling greater power delivery to the load (PDL) of that carrier frequency during non-data transmission periods. Relevant circuits were fabricated using the 180 nm BCD process, and a prototype WPDT system based on these circuits has been successfully developed. Test results show that under a low coupling scenario (17.5 mm coil spacing), the system achieves a PDL of 148 mW while maintaining a DR of 1.1 Mbps with a BER below 10-8. This work fully verifies the system's feasibility and provides an efficient, reliable technical solution for wireless power supply and data transmission in implantable medical devices.


  • Open Access

    LoVoRA: Text-guided and Mask-free Video Object Removal and Addition with Learnable Object-aware Localization

    by Zhihan Xiao, Lin Liu, Yixin Gao, Xiaopeng Zhang, Haoxuan Che, Songping Mai, Qi Tian

    Text-guided video editing, particularly for object removal and addition, remains a challenging task due to the need for precise spatial and temporal consistency. Existing methods often rely on auxiliary masks or reference images for editing guidance, which limits their scalability and generalization. To address these issues, we propose LoVoRA, a novel framework for mask-free video object removal and addition using object-aware localization mechanism. Our approach utilizes a unique dataset construction pipeline that integrates image-to-video translation, optical flow-based mask propagation, and video inpainting, enabling temporally consistent edits. The core innovation of LoVoRA is its learnable object-aware localization mechanism, which provides dense spatio-temporal supervision for both object insertion and removal tasks. By leveraging a Diffusion Mask Predictor, LoVoRA achieves end-to-end video editing without requiring external control signals during inference. Extensive experiments and human evaluation demonstrate the effectiveness and high-quality performance of LoVoRA. https://cz-5f.github.io/LoVoRA.github.io

  • Accurate and Efficient 1D-CNN for Human Activity Recognition Using Millimeter-Wave Point Clouds

    by Anyu Li, Songping Mai

    Human activity recognition (HAR) is critical in applications, such as medical surveillance and contactless human–machine interaction. This letter proposes an accurate and lightweight 1-D convolutional neural network for radar point-cloud-based HAR. The model comprises a lightweight Ghost-PointNet to extract spatial features directly from raw point clouds and a multibranch network to capture multiscale temporal dependencies across the extracted features for classification. Evaluated on the MMActivity dataset, our model achieves 97.82% test accuracy with only 59k parameters, 1.37 MB peak runtime memory usage, and 0.23 giga floating-point operations (GFLOPs). Compared to methods achieving similar accuracy, ours requires up to 99% fewer parameters and 96% smaller input data volumes. By integrating model simplification, ghost convolutions, and depthwise separable convolutions, we maintain high accuracy while drastically reducing computational costs. The ultra-lightweight model is ideal for deployment on resource-constrained edge devices.