arXiv subjects

Siyu You

Publications and source records attributed to Siyu You.

6 recordsLinked to original sources

arXiv2026

Concentrate After Imagination: Text-Conditioned Evidence Grounding for Partially Relevant Video Retrieval

Partially Relevant Video Retrieval (PRVR) retrieves untrimmed videos when queries describe only short moments. Although recent methods improve local representations, uncertainty modeling, and global context, final ranking often still trusts the strongest local response; a coincidentally similar fragment can therefore produce an unsupported peak. We identify this failure as the query-agnostic concentration bottleneck and propose TRACE, a score-level evidence verification operator for PRVR. Given a query and global video registers, TRACE activates query-relevant registers, routes their support to frame-level evidence, and smoothly marginalizes alternative query-to-register-to-frame paths before localized temporal selection. Unlike representation-level feature fusion, TRACE uses this evidence only as a query-conditioned residual calibration of the original local score. On ActivityNet Captions, Charades-STA, and TVR, TRACE achieves the best SumR on all three benchmarks and improves the DreamPRVR backbone by 1.2, 1.1, and 1.5 points, respectively. Ablation, routing-corruption, hard-negative, and cross-backbone transfer analyses support the interpretation that the gains arise from query-conditioned evidence verification rather than a generic score offset.

cs.CV↗

arXiv2026

PhysDox: Benchmarking LLMs on Physical Feasibility Auditing of Physiological Sensing Protocols

Large language models (LLMs) increasingly assist in experimental design, yet fluent protocols often remain physically infeasible. We introduce PhysDox, a physical feasibility auditing benchmark for biomedical protocols comprising a 683-sample expert-curated Gold set and a 5,000-sample Silver set across six sensing domains. We formulate the task as a two-stage evaluation: severity detection classifying protocols as valid, minor, or fatal, followed by the constraint-level diagnosis of fatal violations. Evaluating 6 LLMs across 4 inference strategies yields a peak Stage-1 macro-F1 of only 53.0. Moreover, strong oracle diagnosis collapses during end-to-end evaluation due to correlated cascade errors. Error analysis reveals scaffold bias, where models conflate procedural completeness with physical validity. Consequently, implicit constraints exhibit a 2 times higher miss rate than explicit hardware violations, supported by strong statistical correlation at $\rho{=}0.81$ and $p{<}0.01$. Trace analysis of false negatives exposes a 54%--46% split between attention and judgment failures, ultimately demonstrating that protocol auditing demands calibrated feasibility reasoning rather than factual recall or longer rationales.

cs.HC↗

arXiv2025

A Signal Matrix-Based Local Flaw Detection Framework for Steel Wire Ropes Using Convolutional Neural Networks

Steel wire ropes (SWRs) are critical load-bearing components in industrial applications, yet their structural integrity is often compromised by local flaws (LFs). Magnetic Flux Leakage (MFL) is a widely used non-destructive testing method that detects defects by measuring perturbations in magnetic fields. Traditional MFL detection methods suffer from critical limitations: one-dimensional approaches fail to capture spatial relationships across sensor channels, while multi-dimensional image-based techniques introduce interpolation artifacts and computational inefficiencies. This paper proposes a novel detection framework based on signal matrices, directly processing raw multi-channel MFL signals using a specialized Convolutional Neural Network for signal matrix as input (SM-CNN). The architecture incorporates stripe pooling to preserve channel-wise features and symmetric padding to improve boundary defect detection. Our model achieves state-of-the-art performance with 98.74% accuracy and 97.85% recall. Additionally, it demonstrates exceptional computational efficiency, processing at 87.72 frames per second (FPS) with a low inference latency of 2.6ms and preprocessing time of 8.8ms. With only 1.48 million parameters, this lightweight design supports real-time processing, establishing a new benchmark for SWR inspection in industrial settings.

eess.SP↗

arXiv2025

Local Flaw Detection with Adaptive Pyramid Image Fusion Across Spatial Sampling Resolution for SWRs

The inspection of local flaws (LFs) in Steel Wire Ropes (SWRs) is crucial for ensuring safety and reliability in various industries. Magnetic Flux Leakage (MFL) imaging is commonly used for non-destructive testing, but its effectiveness is often hindered by the combined effects of inspection speed and sampling rate. To address this issue, the impacts of inspection speed and sampling rate on image quality are studied, as variations in these factors can cause stripe noise, axial compression of defect features, and increased interference, complicating accurate detection. We define the relationship between inspection speed and sampling rate as spatial sampling resolution (SSR) and propose an adaptive SSR target-feature-oriented (AS-TFO) method. This method incorporates adaptive adjustment and pyramid image fusion techniques to enhance defect detection under different SSR scenarios. Experimental results show that under high SSR scenarios, the method achieves a precision of 94.73% and a recall of 96.77%. It remains robust under low SSR scenarios with a precision of 94.30% and recall of 97.32%. The overall results show that the proposed method outperforms conventional approaches, achieving state-of-the-art performance. This improvement in detection accuracy and robustness is particularly valuable for handling complex inspection conditions, where inspection speed and sampling rate can vary significantly, making detection more robust and reliable in industrial settings.

eess.IV↗

arXiv2025

DT4ECG: A Dual-Task Learning Framework for ECG-Based Human Identity Recognition and Human Activity Detection

This article introduces DT4ECG, an innovative dual-task learning framework for Electrocardiogram (ECG)-based human identity recognition and activity detection. The framework employs a robust one-dimensional convolutional neural network (1D-CNN) backbone integrated with residual blocks to extract discriminative ECG features. To enhance feature representation, we propose a novel Sequence Channel Attention (SCA) mechanism, which combines channel-wise and sequential context attention to prioritize informative features across both temporal and channel dimensions. Furthermore, to address gradient imbalance in multi-task learning, we integrate GradNorm, a technique that dynamically adjusts loss weights based on gradient magnitudes, ensuring balanced training across tasks. Experimental results demonstrate the superior performance of our model, achieving accuracy rates of 99.12% in ID classification and 90.11% in activity classification. These findings underscore the potential of the DT4ECG framework in enhancing security and user experience across various applications such as fitness monitoring and personalized healthcare, thereby presenting a transformative approach to integrating ECG-based biometrics in everyday technologies.

eess.SP↗

arXiv2025

Improved VMD Based Remote Heartbeat Estimation Utilizing 60GHz mmWave Radar

This study introduces an improved VMD based signal decomposition methodology for non-contact heartbeat estimation using millimeterwave (mmWave) radar. Specifically, we first analyze the signal model of the mmWave radar system. The Variational Mode Decomposition (VMD) integrated with the Newton-Raphson-based optimizer (NRBO) algorithm are sequentially utilized for cardiac mechanic signal (CMS) reconstruction. The estimation accuracy is enhanced by adaptively optimizing the VMD parameters including intrinsic mode functions (IMFs) and penalty factor. Eventually, the experimental results of 18 subjects validate the effectiveness of the proposed method by comparing with three commonly used baselines.

eess.SP↗