The rapid outbreak of COVID-19 has caused serious harm and infected tens of millions of people worldwide. Since there is no specific treatment, wearing masks has become an effective method to prevent the transmission of COVID-19 and is required in most public areas, which has also led to a growing demand for automatic real-time mask detection services to replace manual reminding. However, few studies on face mask detection are being conducted. It is urgent to improve the performance of mask detectors. In this paper, we proposed the Properly Wearing Masked Face Detection Dataset (PWMFD), which included 9205 images of mask wearing samples with three categories. Moreover, we proposed Squeeze and Excitation (SE)-YOLOv3, a mask detector with relatively balanced effectiveness and efficiency. We integrated the attention mechanism by introducing the SE block into Darknet53 to obtain the relationships among channels so that the network can focus more on the important feature. We adopted GIoUloss, which can better describe the spatial difference between predicted and ground truth boxes to improve the stability of bounding box regression. Focal loss was utilized for solving the extreme foreground-background class imbalance. Besides, we performed corresponding image augmentation techniques to further improve the robustness of the model on the specific task. Experimental results showed that SE-YOLOv3 outperformed YOLOv3 and other state-of-the-art detectors on PWMFD and achieved a higher 8.6% mAP compared to YOLOv3 while having a comparable detection speed.

Three-dimensional human pose estimation (3D HPE) from monocular RGB cameras is a fundamental yet challenging task in computer vision, forming the basis of a wide range of applications such as action recognition, metaverse, self-driving, and healthcare. Recent advances in deep learning have significantly propelled the field, particularly with the incorporation of state-space models (SSMs) and diffusion models. However, systematic reviews that comprehensively cover these emerging techniques remain limited. This survey contributes to the literature by providing the first comprehensive analysis of recent innovative approaches, featuring diffusion models and SSMs within 3D HPE. It categorizes and analyzes various techniques, highlighting their strengths, limitations, and notable innovations. Additionally, it provides a detailed overview of commonly employed datasets and evaluation metrics. Furthermore, this survey offers an in-depth discussion on key challenges, particularly depth ambiguity and occlusion issues arising from single-view setups, thoroughly reviewing effective solutions proposed in recent studies. Finally, current applications and promising avenues for future research are highlighted to guide and inspire ongoing innovation in the area, with emerging trends such as integrating large language models (LLMs) to provide semantic priors and prompt-based supervision for improved 3D pose estimation.

Transparent objects, commonly encountered in everyday environments, present significant challenges for 6D pose estimation due to their unique optical properties. The lack of inherent texture and color complicates traditional vision methods, while the transparency prevents depth sensors from accurately capturing geometric details. We propose EBFA-6D, a novel end-to-end 6D pose estimation framework that directly predicts the 6D poses of transparent objects from a single RGB image. To overcome the challenges introduced by transparency, we leverage the high contrast at object boundaries inherent to transparent objects by proposing a boundary feature augmented mechanism. We further conduct a bottom-up feature fusion to enhance the location capability of EBFA-6D. EBFA-6D is evaluated on the ClearPose dataset, outperforming the existing methods in accuracy while achieving an inference speed near real-time. The results demonstrate that EBFA-6D provides an efficient and effective solution for accurate 6D pose estimation of transparent objects.