In complex and crowded scenarios, Multiple Object Tracking (MOT) frequently suffers from severe observation noise and feature degradation induced by frequent target occlusion and motion blur. Existing Tracking-by-Detection (TBD) paradigms typically employ detection confidence thresholds for multi-stage data association, making them highly susceptible to trajectory fragmentation and identity switches (IDSW) among high-quality tracklets. To address these issues, we propose SmoC-Track, a robust MOT approach that dynamically models track confidence to integrate reliable confidence cues as prior constraints into the data association process. First, a multi-dimensional Track Confidence Modeling (TCM) module is proposed to effectively smooth out feature degradation noise during occlusions by fusing long-term historical accumulated states with continuous temporal association features. Second, based on the modeled confidence, we design the Confidence-Height Joint Intersection over Union (CHIoU), a confidence-guided adaptive boundary buffer metric. This mechanism comprises two synergistic components: CIoU adaptively expands the matching search space based on track confidence, whereas HIoU explicitly leverages the invariance prior of bounding box heights in crowded scenes to achieve precise identity decoupling among highly overlapping candidates. Extensive experiments demonstrate that SmoC-Track achieves a HOTA of 66.4% and an IDF1 of 81.8% on the MOT17 test set, as well as 66.9% and 83.1% on the MOT20 test set. These results comprehensively validate its robustness and effectiveness in complex scenarios.
: Existing graph-enhanced sequential recommendation methods typically adopt a unidirectional information flow, in which graph embeddings are injected into the sequential encoder only at the input stage, after which the graph signal is progressively diluted through multiple layers of deep processing. In this paper, the graph signal dilution phenomenon is analyzed systematically across three levels—the input, representation, and prediction layers— and the GSPRec model is proposed to address this issue. The core of GSPRec is the Graph-Sequence Collaborative Injection (GSCI) module, comprising three lightweight components: the Graph Confidence Gate (GCG) controls GCN smoothing via dimension-wise bounded interpolation; the Graph Residual Aggregation (GRA) restores diluted signals through a graph skip connection; and the Graph Collaborative Prediction (GCP) injects collaborative signals into prediction scores via a de-meaned shortcut. GSCI introduces only 194 learnable parameters in total and is equipped with a zero-damage initialization guarantee. The sequential encoder is further enhanced with independent dual Mamba instances and an adaptive path router. Experiments on four benchmark datasets—Food, Movie, Book, and Douban— demonstrate that GSPRec outperforms eight baselines across all 16 evaluation metrics, with relative improvements of 0.61%–3.16% over the strongest baseline and less than 4% additional training time. A layer-wise probing analysis directly confirms that the graph signal is diluted within the sequential encoder and that the GSCI components counteract this loss, while the learned injection strengths adapt across datasets.
Yitao Yang, Peng Wu, Xiaoming Zhang et al.· Computers, Materials & C...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
