event timestamping

Synchronizing football event data with player tracking data remains challenging due to temporal and spatial inaccuracies inherent in manual event annotation; existing approaches rely on error-prone manual localization of events, leading to misalignment. This paper proposes an end-to-end synchronization framework that eliminates the need for manual position annotations by leveraging only player trajectory features. It introduces the first explicit modeling and detection of the termination instants of pass-type events and incorporates a primary-secondary event separation mechanism to mitigate error propagation. The method integrates trajectory-driven feature extraction, temporal alignment algorithms, and event-level decoupling strategies. Evaluated on 2,134 events across three Eredivisie matches, the approach achieves significantly higher synchronization accuracy than state-of-the-art methods, while markedly improving both data completeness and temporal alignment fidelity.

High-quality, publicly available segmentation annotations for dynamic 3D morphological modeling of migrating MDA-MB-231 cells remain scarce, hindering benchmarking and development of accurate 3D live-cell segmentation methods. Method: We introduce the first large-scale, manually curated 3D time-lapse microscopy segmentation dataset with comprehensive spatiotemporal annotations. Annotations integrate Cell Tracking Challenge lineage labels and 2D ground-truth masks to ensure cross-dimensional consistency, validated against automated “silver-standard” segmentations. Inter-annotator variability analysis confirms high reproducibility and fidelity in capturing complex cellular deformations and motility. Contribution/Results: This dataset fills a critical gap in 3D live-cell dynamic segmentation benchmarks. It enables robust training and evaluation of deep learning segmentation models and facilitates quantitative 3D morphodynamic analysis, establishing a new standard for rigorous, data-driven investigation of cellular dynamics.

Discrete event sequences often exhibit both bursty clustering and long inter-event intervals on a standard linear time axis, causing visual clutter and uneven spatial utilization. To address this, we propose a nonlinear time-mapping method based on visual deformation of the time axis. Our approach adaptively compresses low-density temporal regions and stretches high-density ones, directly encoding temporal scale changes as geometric deformations—such as bending and nonuniform scaling—of the time axis itself, thereby enabling intuitive perception of event density. Through crowdsourced graphical perception experiments, we demonstrate that our design significantly outperforms conventional linear and piecewise scaling baselines. Specifically, it reduces event occlusion by 62%, improves user accuracy in judging temporal intervals by 31%, and increases recognition accuracy for periodic and bursty patterns by 27%. This work introduces, for the first time, explicit time-axis deformation as a dedicated temporal-scale encoding mechanism, establishing a novel paradigm for event sequence visualization.

Video large language models (Video LLMs) suffer from insufficient accuracy in zero-shot temporal grounding, hindering their practical deployment in video understanding and editing. To address this, we propose TimeLLM: the first framework to explicitly inject timestamp knowledge into visual tokens; it introduces absolute time embeddings to mitigate semantic drift across long temporal spans and designs a lightweight, slot-based frame compression mechanism that preserves critical temporal structure while reducing computational overhead. Leveraging the high-quality, re-annotated dataset VTG-IT-120K, TimeLLM significantly outperforms existing Video LLMs across multiple temporal grounding benchmarks. It achieves more robust and precise zero-shot temporal localization, enabling general-purpose, cross-task video interaction without task-specific fine-tuning.

Existing cell tracking methods suffer from significant limitations in long-term consistency, mitosis modeling, and lineage reconstruction accuracy, failing to meet biologically grounded lineage analysis requirements. This paper proposes a mitosis-aware multi-hypothesis assignment framework that introduces, for the first time, test-time augmented motion estimation with quantified uncertainty, and explicitly incorporates biological priors into the assignment cost function. By jointly modeling probability densities and solving the optimal assignment under biologically constrained optimization, our method achieves state-of-the-art performance across nine benchmark datasets. It improves biology-inspired metrics by approximately 6×, and—critically—uncovers, for the first time, statistically significant correlations between motion uncertainty and cellular behaviors (e.g., mitotic timing, migration patterns). This establishes a new paradigm for interpretable and verifiable lineage reconstruction.

Existing cell tracking methods struggle with transient, time-varying fluorescent signals (e.g., circadian rhythms) in multi-channel microscopy videos and fail to reliably detect critical cellular events such as division and apoptosis, thereby limiting single-cell dynamic analysis. To address this, we propose a lightweight deep learning framework that performs multi-object matching directly in the detection embedding space. Our approach integrates a Transformer to model long-range temporal dependencies, employs multi-stage full-bounding-box matching for enhanced robustness, and incorporates Kalman filter-based interpolation to mitigate signal fluctuations. The method operates end-to-end without handcrafted features, generating complete, lineage-aware cell trajectories. Evaluated across diverse experimental conditions, it significantly improves tracking accuracy and precisely captures mitotic and apoptotic events. We successfully applied it to assess single-cell responses to chemotherapy, enabling quantitative, interpretable analysis of cancer cell dynamics.

This work addresses the challenge of users struggling to pinpoint specific moments in meeting discussions based solely on content. To overcome this, the paper proposes a novel approach that reframes timestamp prediction as a constrained candidate selection task. Instead of directly generating timestamps, large language models such as Mistral-7B-Instruct are guided to select the most relevant segment from a set of retrieved, timestamped meeting excerpts, thereby avoiding unsupported or invalid predictions. Integrating retrieval-augmented generation (RAG) with a constrained selection mechanism, the method demonstrates significant improvements on a dataset of 200 municipal meetings and 420 queries: Recall@5 increases from 31.9% to 50.0%, mean absolute error decreases to 761 seconds, and the number of valid outputs rises from 373 to 419, substantially enhancing both accuracy and reliability in temporal localization.

Existing data discovery methods largely overlook the temporal dynamics of data relationships, particularly hindering timeliness-aware retrieval when explicit temporal metadata is absent. To address this, we propose “temporal-effective data discovery”—a novel paradigm introducing the first time-integrated data discovery system. Its core comprises four techniques: version discovery, temporal lineage inference, change-log synthesis, and time-aware querying. Crucially, the system models semantic evolution of data over time without relying on original timestamps, enabling dynamic lineage tracing and version-level precise retrieval. Experimental evaluation on enterprise and public data lakes demonstrates significant improvements in accuracy and practicality for timeliness-sensitive queries. Our work establishes a new research direction in data discovery centered on temporal evolution, advancing both theoretical foundations and real-world applicability.

Fluorescence microscopy videos are often compromised by noise, temporal variability, and signal oscillations, hindering accurate analysis of dynamic biological processes. This work proposes an interpretable, end-to-end computational framework that, for the first time, integrates multi-temporal image registration, feature alignment, and cross-domain interpretable visual algorithms to efficiently compress dynamic video sequences into a single high-quality image while preserving critical biological structures. Experiments on a complex dataset of cardiomyocyte monolayers demonstrate that the proposed method increases the average number of detected cells by 44% compared to existing approaches, significantly enhancing both image quality and downstream segmentation performance.

This study addresses the pervasive issue of logical inconsistencies in football event stream data—such as misplaced or missing events—that severely compromise analytical reliability. To this end, we propose VERSA, a novel framework that introduces, for the first time, a systematic validation mechanism grounded in state transition modeling to automatically detect and rectify anomalous event sequences. By integrating event stream validation algorithms with data repair techniques, VERSA substantially enhances both data integrity and cross-provider consistency. Experimental evaluation on K League 1 data reveals that 18.81% of events exhibit logical errors; after VERSA processing, cross-provider consistency improves markedly, and performance on downstream VAEP-based player contribution assessment tasks shows significant gains.