Takahashi HIROKI (髙橋 裕樹)

The sparsest packing problem emerges in manufacturing multi-hole extrusion dies to obtain small and precise products in the automotive, aviation, food, and medical industries. The goal is to maximize the minimum Euclidean distance between the objects and between the objects to the boundaries of the container. Additionally, this task might be subject to balancing constraints that determine that the deviation of gravity center of the system should stay within a threshold. We present a novel custom environment that encompasses the constraints present in this task. We experiment with the proposed environment using Proximal Policy Optimization to assess the applicability of reinforcement learning for the sparsest packing problem with circular objects in a circular container. Our results indicate that the proposed agent learns efficiently, demonstrating promising results in both finding feasible solutions and optimizing the placement of objects. Our approach not only shows the potential of reinforcement learning for solving the sparsest packing problem but also provides insights into its effectiveness in environments with complex spatial and balancing constraints.

Video surveillance systems are a key tool for the identification of anomalous events, but they still rely heavily on human analysis, which limits their efficiency. Current video anomaly detection models aim to automatically detect such events. However, most of them provide only a binary classification (normal or anomalous) and do not identify the specific type of anomaly. Although recent proposals address anomaly classification, they typically require full video analysis, making them unsuitable for online applications. In this work, we propose a modular framework for the joint detection and classification of anomalies, designed to operate on individual clips within continuous video streams. The architecture integrates interchangeable modules (feature extractor, detector, and classifier) and is adaptable to both offline and online scenarios. Specifically, we introduce a multi-category classifier that processes only anomalous clips, enabling efficient clip-level classification. Experiments conducted on the UCF-Crime dataset validate the effectiveness of the framework, achieving 74.77% clip-level accuracy and 58.96% video-level accuracy, surpassing prior approaches and confirming its applicability in real-world surveillance environments.

Hand gesture recognition (HGR) heavily relies on high-quality annotated datasets. However, annotating hand landmarks in video sequences is a time-intensive challenge. In this work, we introduce IPN HandS, an enhanced version of our IPN Hand dataset, which now includes approximately 700,000 hand skeleton annotations and corrected gesture boundaries. To generate these annotations efficiently, we propose a novel annotation tool that combines automatic detection, inter-frame interpolation, copy–paste capabilities, and manual refinement. This tool significantly reduces annotation time from 70 min to just 27 min per video, allowing for the scalable and precise annotation of large datasets. We validate the advantages of the IPN HandS dataset by training a lightweight LSTM-based model using these annotations and comparing its performance against models trained with annotations from the widely used MediaPipe hand pose estimators. Our model achieves an accuracy that is 12% higher than the MediaPipe Hands model and 8% higher than the MediaPipe Holistic model. These results underscore the importance of annotation quality in training generalization and overall recognition performance. Both the IPN HandS dataset and the annotation tool will be released to support reproducible research and future work in HGR and related fields.

Human Pose Estimation (HPE) remains challenging due to scale variation, occlusion, and high computational costs. Standard methods often struggle to capture detailed spatial information when keypoints are obscured, and they typically rely on computationally expensive deconvolution layers for upsampling, making them inefficient for real-time or resource-constrained scenarios. We propose AMFACPose (Attentive Multi-scale Features with Adaptive Context PoseResNet) to address these limitations. Specifically, our architecture incorporates Coordinate Convolution 2D (CoordConv2d) to retain explicit spatial context, alleviating the loss of coordinate information in conventional convolutions. To reduce computational overhead while maintaining accuracy, we utilize Depthwise Separable Convolutions (DSCs), separating spatial and pointwise operations. At the core of our approach is an Adaptive Feature Pyramid Network (AFPN), which replaces costly deconvolution-based upsampling by efficiently aggregating multi-scale features to handle diverse human poses and body sizes. We further introduce Dual-Gate Context Blocks (DGCBs) that refine global context to manage partial occlusions and cluttered backgrounds. The model integrates Squeeze-and-Excitation (SE) blocks and the Spatial–Channel Refinement Module (SCRM) to emphasize the most informative feature channels and spatial regions, which is particularly beneficial for occluded or overlapping keypoints. For precise keypoint localization, we replace dense heatmap predictions with coordinate classification using Multi-Layer Perceptron (MLP) heads. Experiments on the COCO and CrowdPose datasets demonstrate that AMFACPose surpasses the existing 2D HPE methods in both accuracy and computational efficiency. Moreover, our implementation on edge devices achieves real-time performance while preserving high accuracy, confirming the suitability of AMFACPose for resource-constrained pose estimation in both benchmark and real-world environments.

This paper presents SqueezeMaskNet, a lightweight convolutional neural network designed for real-time recognition of proper and improper mask usage. The model classifies four categories: masks worn correctly, masks covering only the mouth, masks not covering, and no mask. SqueezeMaskNet integrates seamlessly with existing face detection systems, removing the need for retraining. We propose using Fire modules for efficiency, along with attention mechanisms like efficient channel attention (ECA) and squeeze-and-excitation (SE) blocks for improved feature refinement. SqueezeMaskNet achieved 96.7% accuracy on the challenging FineFM test set and ran at 297 FPS on a GPU and up to 96 FPS on edge devices like a Jetson Orin NX. We also introduced ImproperTFM, a subset of real-world images focusing on improper mask usage, which enhanced the model accuracy when combined with FineFM data. Comparative experiments demonstrated SqueezeMaskNet’s superior performance, efficiency, and adaptability compared to MobileNet and EfficientNet, making it a practical solution for mask-wearing recognition across various devices and settings.