—Human beings rely primarily on vision to perceive and interact with the external world, with approximately 80% of sensory information input through the visual system. This visual dominance makes the question of "where an individual is looking" not only a key to understanding attention distribution and information processing mechanisms but also a critical factor in optimizing decision-making efficiency and learning outcomes. However, traditional methods for analyzing gaze-related behaviors — such as manual behavioral observation and self-reported evaluation— suffer from inherent limitations: be havioral observation relies on subjective judgment of observers, often missing subtle gaze shifts and failing to achieve real-time tracking; self-evaluation is prone to memory biases and social desirability effects, leading to deviations between reported and actual gaze patterns. These drawbacks highlight the need for a more objective and precise alternative.Gaze estimation, which infers an individual’s visual attention and behavioral intentions by recording and analyzing the spatial position, movement trajectory, and dynamic changes of the eyeball, emerges as an ideal solution. This technology is broadly categorized into model-based (relying on geometric eye models) and appearance-based (using facial/ocular image features) approaches, with appearance-based methods gaining traction due to their non-intrusiveness. Nevertheless, current appearance-based gaze estimation still faces two major challenges: (1) individual differences, such as variations in eye shape, pupil size, eyelid structure, and the presence of glasses, which disrupt consistent feature extraction; (2) environmental interference, including variable lighting, partial facial occlusion, and dynamic head poses, which reduce estimation accuracy. To address these issues, this paper proposes RTACM-Net, a novel gaze estimation network architecture that integrates the strengths of Vision Transformer (ViT) with a multi-scale feature fusion mechanism. Specifically, RTACM-Net employs a lightweight convolutional module to extract local fine-grained features of the ocular region, while leveraging ViT’s multi-head attention mechanism to capture global contextual relationships. This dual-branch design enables the network to balance local feature precision and global context awareness, thereby mitigating the impact of individual differences and environmental noise.Extensive experiments were conducted on two benchmark datasets: MPIIFaceGaze (a large-scale dataset focusing on indoor controlled environments with 21 subjects) and Gaze360 (a challenging dataset covering diverse outdoor/indoor scenes, variable lighting, and large head-pose variations with over 100 subjects). The results show that RTACM-Net : on MPIIFaceGaze, it achieves an average angular error (MAE) of 3.72° ; on Gaze360, it achieves an MAE of 10.46°, Gaze360-Net (11.40°) by 0.94°. These results demonstrate the robustness of RTACM-Net in handling variable individual characteristics and complex environmental conditions. Its practical potential extends to multiple fields: in augmented reality (AR), it can enable adaptive interface rendering; in autonomous driving, it supports dual-task monitorin; in human-robot interaction, it facilitates intuitive service triggering.
—This paper presents an improved version of the DeepLabV3+ network to address issues such as large parameter count, difficulties in mobile deployment, limited receptive field, and insufficient utilization of low-level semantic information in existing deep learning semantic segmentation networks. The main enhancement approach is as follows: we utilize the lightweight MobileNetV2 as the backbone feature extraction network, while an improved multi-scale atrous convolution module (AS-ASPP) and convolutional block attention mechanism (CBAM) are introduced. Tests conducted on the PASCAL VOC 2012 dataset demonstrate that the optimized model retains merely around one-tenth the parameters of the original network, while attaining superior segmentation precision and computational effectiveness. Specifically, it reaches a mIoU of 73.21% and a Precision of 80.56%, with the training time reduced by approximately 50% and the inference speed significantly improved.
—The rapid advancement of mobile technologies has led to increasingly powerful and feature-rich devices, yet this progress has also intensified the challenge of managing energy consumption effectively. Power optimization has therefore become a critical focus in mobile operating systems (OS), aiming to balance performance, functionality, and energy efficiency. As mobile devices integrate more complex hardware components and resource-intensive applications, ensuring sustainable power usage has become essential for improving battery life, user experience, and environmental sustainability. This research explores the fundamental question: How can mobile operating systems intelligently manage hardware and software resources to minimize power consumption without compromising performance or usability? To address this, the study examines key power optimization strategies and mechanisms integrated within modern mobile OS architectures, including Dynamic Voltage and Frequency Scaling (DVFS), power-aware CPU scheduling, Doze and App Standby modes, adaptive display and sensor management, and network optimization. The research also investigates the role of advanced techniques such as context-aware power management and machine learning-based predictive models in achieving dynamic, intelligent energy control. Using tools like Trepn Profiler, PowerTutor, and Android Battery Historian, the study evaluates how power consumption patterns can be analyzed and optimized in real time. The findings reveal that combining hardware-level techniques (like voltage scaling and clock gating) with software-level optimizations (such as adaptive scheduling and contextual awareness) results in significant energy savings while maintaining user satisfaction. Furthermore, the study highlights emerging challenges, including the trade-offs between performance and energy efficiency, the integration of AI for predictive optimization, and the need for sustainability across the device lifecycle. Ultimately, this research demonstrates that power optimization in mobile operating systems is not merely a technical requirement but a cornerstone of sustainable computing. Through intelligent power management, future mobile OSs can achieve greater efficiency, extended device longevity, and reduced environmental impact aligning technological innovation with eco-efficiency and user-centric design.
—To address the challenge of accurate gaze estimation in unconstrained environments susceptible to various interfering factors, this paper proposes AG-HybridNet, an end-to-end gaze estimation model integrating a dual-branch architecture combining CNN and Transformer components. The model employs Swin Transformer as the backbone for global feature extraction while incorporating an enhanced CNN branch dedicated to local feature capture. We introduce the TDConv-Block, which replaces standard convolution with partial convolution integrated with reparameterization technique, significantly reducing computational load and memory access while forming a T-shaped receptive field focused on central facial regions. Additionally, we design Efficient Additive Attention (ED-Attention) that effectively resolves the computational bottleneck in long-sequence processing for Transformers by reconstructing the computational workflow. Comprehensive experiments on MPIIFaceGaze and Gaze360 datasets validate the model's effectiveness. Experimental results demonstrate that AG-HybridNet achieves mean angular errors of 3.72° and 10.82° on MPIIFaceGaze and Gaze360 datasets respectively. Comparative studies with other mainstream 3D gaze estimation methods confirm that our network model can accurately estimate 3D gaze directions while reducing computational complexity.
—Small object detection remains a formidable challenge in computer vision, primarily because conventional models like SSD suffer from two critical limitations: weak semantic information in shallow feature maps and a mismatch between the receptive field and the actual size of small targets. To address these deficiencies, this paper introduces Lite-RFB SSD, an innovative architecture that strategically integrates a lightweight Receptive Field Block (RFB) module into the SSD framework. This module is meticulously reconstructed using depthwise separable convolutions and channel pruning techniques, resulting in a remarkable 62% reduction in parameters. By embedding this optimized module into the shallow conv4_3 layer, the model preserves high-resolution features crucial for small object detection while significantly enhancing computational efficiency. Experimental validation on the PASCAL VOC dataset demonstrates that Lite-RFB SSD achieves an average precision for small objects (APs) of 22.9%, a substantial 4.2% improvement over the original SSD. Furthermore, it operates at an impressive 28 FPS on edge devices, establishing a superior balance between accuracy and efficiency that outperforms competing methods such as standard RFB and MobileNet-SSD.
