Zhihao Su

Multimodal multi-object tracking has emerged as a crucial task in autonomous driving, aiming to perceive and track dynamic objects through the integration of visual, auditory, and linguistic cues. However, this task faces two major challenges: insufficient multimodal fusion, particularly the loss of frequency-domain information during spatial-only processing, and weak temporal semantic consistency, which often leads to tracking drift across frames. To address these issues, previous studies have explored spatial–frequency fusion networks and contrastive learning frameworks that align heterogeneous features across modalities. While these methods have achieved notable progress, they still suffer from limited adaptability to complex multimodal interactions and unstable long-term dependencies. To overcome these limitations, we propose SynTrack, a synergistic multimodal tracking framework that unifies frequency-domain fusion and contrastive temporal learning. SynTrack introduces two key components: a Multi-scale Cross-modal Frequency Fusion Module (MCFM) to enhance cross-modal representation through dual-domain attention, and a Multimodal Contrastive Tracking Learning (MCTL) strategy to enforce temporal coherence and semantic alignment across modalities. Extensive experiments on newly curated multimodal and text-guided benchmarks demonstrate that SynTrack achieves superior tracking accuracy, stability, and generalization under diverse environmental conditions, validating its effectiveness and scalability for real-world autonomous driving scenarios.

Keypoint-based deep learning detectors have proven highly effective in object detection tasks by predicting specific keypoints to determine object classification and location. Examples include Corner-Net, CenterNet, ExtremeNet, RetinaNet, FCOS, and ObjectBox. Despite their strengths, these methods are particularly susceptible to class imbalance, which can result in poor detection performance for less frequent classes. Popular solutions such as hard sampling, soft sampling, and sampling-free methods have been proposed to tackle this issue. However, these approaches often have inherent limitations, including sensitivity to hyperparameters and neglecting the gradient dynamics of the loss function. To address these challenges, this paper proposes Adaptive Focal Loss (AFL), which combines the strengths of Focal Loss (FL) and Class-Balanced Loss (CBL) while introducing an Adaptive Gradient Function. This function is specifically designed to mitigate the impact of large gradients during the early stages of training. Extensive experiments demonstrate that AFL significantly outperforms classical sampling methods across key metrics on the MS COCO, LVIS, and PASCAL VOC datasets. Notably, on the LVIS dataset, AFL achieves an average improvement of over 3% AP_r compared to the baseline Focal Loss across multiple keypoint-based detectors, showcasing its effectiveness in enhancing rare class accuracy. Furthermore, AFL delivers substantial improvements in the AP_0.5~0.95 metric, surpassing the baseline Focal Loss by an average of over 2%, 1%, and 0.2% on the LVIS, MS COCO, and PASCAL VOC datasets, respectively. This highlights AFL's capability to enhance the overall accuracy of foreground objects. By providing a more balanced and robust solution to class imbalance, AFL demonstrates superior performance in challenging scenarios, making it a valuable advancement in keypoint-based detection.

Region proposal-based detectors, such as Region-Convolutional Neural Networks (R-CNNs), Fast R-CNNs, Faster R-CNNs, and Region-Based Fully Convolutional Networks (R-FCNs), employ a two-stage process involving region proposal generation followed by classification. This approach is effective but computationally intensive and typically slower than proposal-free methods. Therefore, region proposal-free detectors are becoming popular to balance accuracy and speed. This paper proposes a proposal-free, fully convolutional network (PF-FCN) that outperforms other state-of-the-art, proposal-free methods. Unlike traditional region proposal-free methods, PF-FCN can generate a “box map” based on regression training techniques. This box map comprises a set of vectors, each designed to produce bounding boxes corresponding to the positions of objects in the input image. The channel and spatial contextualized sub-network are further designed to learn a “box map”. In comparison to renowned proposal-free detectors such as CornerNet, CenterNet, and You Look Only Once (YOLO), PF-FCN utilizes a fully convolutional, single-pass method. By reducing the need for fully connected layers and filtering center points, the method considerably reduces the number of trained parameters and optimizes the scalability across varying input sizes. Evaluations of benchmark datasets suggest the effectiveness of PF-FCN: the proposed model achieved an mAP of 89.6% on PASCAL VOC 2012 and 71.7% on MS COCO, which are higher than those of the baseline Fully Convolutional One-Stage Detector (FCOS) and other classical proposal-free detectors. The results prove the significance of proposal-free detectors in both practical applications and future research.

Deep learning models have shown great promise in diagnosing skeletal fractures from X-ray images. However, challenges remain that hinder progress in this field. Firstly, a lack of clear definitions for recognition, classification, detection, and localization tasks hampers the consistent development and comparison of methodologies. The existing reviews often lack technical depth or have limited scope. Additionally, the absence of explainable facilities undermines the clinical application and expert confidence in results. To address these issues, this comprehensive review analyzes and evaluates 40 out of 337 recent papers identified in prestigious databases, including WOS, Scopus, and EI. The objectives of this review are threefold. Firstly, precise definitions are established for the bone fracture recognition, classification, detection, and localization tasks within deep learning. Secondly, each study is summarized based on key aspects such as the bones involved, research objectives, dataset sizes, methods employed, results obtained, and concluding remarks. This process distills the diverse approaches into a generalized processing framework or workflow. Moreover, this review identifies the crucial areas for future research in deep learning models for bone fracture diagnosis. These include enhancing the network interpretability, integrating multimodal clinical information, providing therapeutic schedule recommendations, and developing advanced visualization methods for clinical application. By addressing these challenges, deep learning models can be made more intelligent and specialized in this domain. In conclusion, this review fills the gap in precise task definitions within deep learning for bone fracture diagnosis and provides a comprehensive analysis of the recent research. The findings serve as a foundation for future advancements, enabling improved interpretability, multimodal integration, clinical decision support, and advanced visualization techniques.