WKWataru Kameyama

Wataru Kameyama

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Wataru Kameyama has a total of 18 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Structural Relation Multi-Class Token Transformer for Weakly Supervised Semantic Segmentation

    by Dingjie Peng, Wataru Kameyama

    Weakly Supervised Semantic Segmentation (WSSS) aims to train models to identify and delineate objects within an image using limited training data such as image-level labels. While recent works mainly focus on exploring class-specific knowledge to improve the quality of class activation maps, we contend that relying solely on this approach within a non-hierarchical architecture fails to adequately capture the structural relationships within images. Drawing inspiration from fully supervised semantic segmentation designs, which use hierarchical multi-scale feature maps for predicting the dense masks, we propose a novel architecture that integrates a Structural Relation Multi-class Token Transformer (SR-MCT) with WSSS. This model employs multi-scale structural tokens, generated by a Spatial Prior Module (SPM), which interact not only with patch tokens to encode structural relations, but also with multi-class tokens to integrate class-specific knowledge into complex structural embeddings. The proposed Structural Relation Multi-class Token Attention effectively builds long-range dependencies among structural tokens, patch tokens, and multi-class tokens simultaneously. Experimental results and ablation studies on PASCAL VOC 2012 and MS COCO 2014 demonstrate that our proposed SR-MCT can enhance baseline performance and outperform other state-of-the-art methods.


  • Open Access

    AI-Based W-Band Suspicious Object Detection System for Moving Persons: Two-Stage Walkthrough Configuration and Recognition Optimization

    by Zheng Wen, Keping Yu, Xin Qi, Toshio Sato, San Hlaing Myint, Kazuhiko Tamesue, Yutaka Katsuyama, Hironori Dobashi, Yasushi Murakami, Ikuo Koyama, et al.

    In recent years, terrorist attacks have been spreading worldwide and become a public hazard to human society. The suspicious object detection system is an effective way to prevent terrorist attacks in public places. However, traditional systems face two main challenges: First, they need to conduct security checks at the entrance one by one, which leads to crowding; second, they rely heavily on screeners’ ability to understand security images, which can easily lead to misjudgment. To address these issues, we propose an AI-based W-band suspicious object detection system for moving persons that can perform a two-stage walkthrough screening for suspicious objects in an open area to maintain high throughput. The 1st screening uses millimeter wave radar and cameras to automatically screen suspects who may have concealed suspicious objects in an open area. The 2nd screening involves security personnel using a hybrid imager with active and passive imaging capabilities to identify the specific suspicious objects carried by the suspect. Convolutional neural network (CNN) based artificial intelligence (AI) technology will be used to improve the accuracy and speed of suspicious object detection. We performed an experiment to validate the proposed system. The usability and safety of the system are demonstrated by recognition rate (aka accuracy rate) or both recall and precision rate. In addition, in the process of improving the suspicious object recognition rate by AI techniques, we use generative adversarial network to help build a suspicious object database and successfully validate the effectiveness of the method and the factors affecting the suspicious object recognition rate to optimize the system.


  • Open Access

    Classifying High Quality Photographs by Creative Exposure Themes

    by Supheakmungkol Sarin, Wataru Kameyama

    In this paper, we propose to utilize contextual camera setting parameters at the time of capture to perform the classification task of high quality photographs. With supervised machine learning algorithm, we build a model that can classify high quality photographs into six creative exposure themes which are commonly known and used by the professional photographers. Our experiments give us an encouraging result.

  • Hybrid Approach for Khmer Unknown Word POS Guessing

    by Chenda Nou, Wataru Kameyama

    New words are being created everyday and the lexicon is not large enough to cover all the words, unknown words become a serious problem in part-of-speech tagging. This paper presents a hybrid approach to handle the unknown word problem in Khmer part-of-speech tagging. The hybrid approach combined of rule-based model and trigram model makes use of both internal structure of the word and surrounding contextual information to predict the part-of-speech of unknown words. The proposed approach achieves 88.9% and 78.2% of accuracy on training and test data respectively.


  • Exploiting Users' Personal and Public Information for Personal Photo Annotation

    by Supheakmungkol Sarin, Toshinori Nagahashi, Tadashi Miyosawa, Wataru Kameyama

    Annotation is the core item to enable efficient access to media. Thus, automatic annotation is a highly challenging problem. We present a novel solution to personal digital photograph annotation problem which annotates photo based on users' personal, and public information. We use timestamps of photo together with exact location name as pivots to search for more contextual metadata from users' emails, schedules, web browsing histories and public information repositories such as online encyclopedia and news Web sites. A notable contribution of our approach is that it can effectively leverage the background information of users' daily activities and their ambient environment to semi-automatically annotate their photographs. Experimental results conducted with 10 subjects and a total of 313 photos show that we can reduce 33% of annotation time over manual annotation and we also obtain very encouraging accuracy results.


  • A Proposal of Ontology-based Health Care Information Extraction System: VnHIES

    by Tran Quoc Dung, Wataru Kameyama

    This paper presents an ontology-based health care information extraction system - VnHIES. In the system, we develop and use two effective algorithms called "semantic elements extracting algorithm" and "new semantic elements learning algorithm" for health care semantic words extraction and ontology enhancement. The former algorithm will extract concepts (Cs), descriptions of concepts (Ds), pairs of concept and description(C-D) and Names of diseases (Ns) in health care information domain from Web pages. Those extracted semantic elements are used by latter algorithm that will render suggestions in which might contain new semantic elements for later use by domain users to enrich ontology. After extracting semantic elements, a "document weighting algorithm" is applied to get summary information of document with respect to all extracted semantic words and then to be stored in knowledge base which contains ontology and database to be used later in other applications. Our experiment results show that the approach is very optimistic with high accuracy in semantic extracting and efficiency in ontology upgrade. VnHIES can be used in many health care information management systems such as medical document classification, health care information retrieval system. VnHIES is implemented in Vietnamese language.


  • On Automatic Contextual Metadata Generation for Personal Digital Photographs

    by Supheakmungkol Sarin, Toshinori Nagahashi, Tadashi Miyosawa, Wataru Kameyama

    The advancements of technology have been making digital cameras more and more affordable for everyone. Consequently, digital photographs have grown exponentially this last decade and have been making it beyond the abilities of people to easily manage their own photo collections. Annotation is one of the solutions to solve this burden. Therefore, we propose a semi-automatic annotation engine with contextual metadata generation. Photos with geo-referential information such as Global Positioning System (GPS) are the current tendency. Using the exact location information given by GPS, together with timestamps, our novel engine semi-automatically generates contextual metadata for each photo from different sources of information namely the public news and users' personal information. We have designed, implemented and tested our system.