Propagation and association tasks in Multi-Object Tracking (MOT) play a pivotal role in accurately linking the trajectories of moving objects. Recently, modern deep learning models have been addressing these tasks by introducing fragmented solutions for each different problem such as appearance modeling, motion modeling, and object associations. To bring unification in the MOT task, we introduce a pixel-guided approach to efficiently build the joint-detection and tracking framework for multi-object tracking. Specifically, the up-sampled multi-scale features from consecutive frames are queued to detect the object locations by using a transformer–decoder, and per-pixel distributions are utilized to compute the association matrix according to object queries. Additionally, we introduce a long-term appearance association on track features to learn the long-term association of tracks against detections to compute the similarity matrix. Finally, a similarity matrix is jointly integrated with the Byte-Tracker resulting in a state-of-the-art MOT performance. The experiments with the standard MOT15 and MOT17 benchmarks show that our approach achieves significant tracking performance.

In the landscape of Vishing calls, it's crucial to determine if users are unknowingly falling prey to fraudsters' tactics, potentially leading to fraud. This paper introduces the Multimodal Vishing Threat Detection (MmVTD) framework, designed to counter Vishing threats and issue alerts on potential fraudulent activities leveraging three key data modalities: call transcripts, sequences of mobile screenshots, and extracted Optical Character Recognition(OCR) text. The novelty of our approach lies in utilization of transformer encoders to build the learnable context for each modality, with a decoder generating warning content while an MLP head classifies the current call as Vishing or Safe at any given instance. This innovative architecture enhances the model's ability to analyze the diverse data sources and generate effective warning in real-time. Furthermore, the MmVTD framework offers robust protection through multimodal analysis, providing real-time alerts when user action suggest susceptibility fraud. Finally, the MmVTD model achieved an impressive 94.44% accuracy in identifying Vishing and Safe calls on a dedicated dataset. Additionally, it attained a BLEU score of 0.583 in generating cautionary messages, effectively deterring users from potential harmful actions.