TOHOKU UNIVERSITY Researchers

Details of the Researcher

Home

日本語 English

Asako Kanezaki

Section

Graduate School of Information Sciences

Job title

Professor

Degree

researchmap

https://researchmap.jp/kanezaki

J-GLOBAL ID

201801017711602320

Profile

2008年3月東京大学工学部卒業。2010年3月同大学院情報理工学系研究科修士課程修了。2010年4月より日本学術振興会特別研究員（DC1）。2013年3月同大学院同研究科博士課程修了、博士（情報理工学）。（株）東芝研究開発センター正規職員、同大学院同研究科助教を経て、2016年4月より産業技術総合研究所人工知能研究センター勤務。2020年4月より東京工業大学准教授。2026年4月より東北大学教授、現在に至る。機械学習を用いた三次元物体認識、物体検出、ロボットビジョン、Embodied AIの研究に従事。IEEE RAS Japan Chapter Young Award、 PRMU研究奨励賞、船井研究奨励賞、RSJ研究奨励賞、日本機械学会奨励賞（研究）を受賞。

Research History 1

2026/04 - Present

Tohoku University　Graduate School of Information Sciences　Professor

Research Areas 1

Informatics / Intelligent informatics /

Awards 5

日本機械学会奨励賞（研究）

2021/03　日本機械学会　深層学習を用いた３次元物体認識の研究
Young Investigation Excellence Award

2015/09　The Robotics Society of Japan
Encouraging Prize of Funai Foundation for Information Technology

2014/04　Funai Foundation for Information Technology
Encouraging Prize

2012/09　IEICE Technical Committee on Pattern Recognition and Media Understanding (PRMU)
IEEE Robotics and Automation Society Japan Chapter Young Award (ICRA 2010)

2010/05　IEEE　High-speed 3D Object Recognition Using Additive Features in A linear Subspace

Papers 63

Object Memory Transformer for Object Goal Navigation.

Rui Fukushima, Kei Ota, Asako Kanezaki, Yoko Sasaki, Yusuke Yoshiyasu

CoRR　abs/2203.14708　2022

DOI： 10.48550/arXiv.2203.14708 　
CTRMs: Learning to Construct Cooperative Timed Roadmaps for Multi-agent Path Planning in Continuous Spaces.

Keisuke Okumura 0001, Ryo Yonetani, Mai Nishimura, Asako Kanezaki

CoRR　abs/2201.09467　2022
CTRMs: Learning to Construct Cooperative Timed Roadmaps for Multi-agent Path Planning in Continuous Spaces.

Keisuke Okumura 0001, Ryo Yonetani, Mai Nishimura, Asako Kanezaki

AAMAS　972-981　2022
Publisher: International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS)
OPIRL: Sample Efficient Off-Policy Inverse Reinforcement Learning via Distribution Matching.

Hana Hoshino, Kei Ota, Asako Kanezaki, Rio Yokota

CoRR　abs/2109.04307　2021
Training Larger Networks for Deep Reinforcement Learning.

Kei Ota, Devesh K. Jha, Asako Kanezaki

CoRR　abs/2102.07920　2021
Path Planning using Neural A* Search.

Ryo Yonetani, Tatsunori Taniai, Mohammadamin Barekatain, Mai Nishimura, Asako Kanezaki

Proceedings of the 38th International Conference on Machine Learning(ICML)　12029-12039　2021
Publisher: PMLR
Deep Reactive Planning in Dynamic Environments.

Kei Ota, Devesh K. Jha, Tadashi Onishi, Asako Kanezaki, Yusuke Yoshiyasu, Yoko Sasaki, Toshisada Mariyama, Daniel Nikovski

CoRR　abs/2011.00155　2020
Incremental multi-view object detection from a moving camera.

Takashi Konno, Ayako Amma, Asako Kanezaki

MMAsia 2020: ACM Multimedia Asia(MMAsia)　4-7　2020
Publisher: ACM
DOI： 10.1145/3444685.3446257 　
Efficient Exploration in Constrained Environments with Goal-Oriented Reference Path.

Kei Ota, Yoko Sasaki, Devesh K. Jha, Yusuke Yoshiyasu, Asako Kanezaki

IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS)　6061-6068　2020
Publisher: IEEE
DOI： 10.1109/IROS45743.2020.9341620 　
Deep Reactive Planning in Dynamic Environments.

Kei Ota, Devesh K. Jha, Tadashi Onishi, Asako Kanezaki, Yusuke Yoshiyasu, Yoko Sasaki, Toshisada Mariyama, Daniel Nikovski

4th Conference on Robot Learning(CoRL)　1943-1957　2020
Publisher: PMLR
Unsupervised Learning of Image Segmentation Based on Differentiable Feature Clustering.

Wonjik Kim, Asako Kanezaki, Masayuki Tanaka 0001

IEEE Transactions on Image Processing　29　8055-8068　2020

DOI： 10.1109/TIP.2020.3011269 　
Path Planning using Neural A* Search. Peer-reviewed

Ryo Yonetani, Tatsunori Taniai, Mohammadamin Barekatain, Mai Nishimura, Asako Kanezaki

CoRR　abs/2009.07476　2020
Unsupervised Learning of Image Segmentation Based on Differentiable Feature Clustering Peer-reviewed

Wonjik Kim, Asako Kanezaki, Masayuki Tanaka

IEEE TRANSACTIONS ON IMAGE PROCESSING　29　8055-8068　2020

DOI： 10.1109/TIP.2020.3011269 　

ISSN： 1057-7149

eISSN： 1941-0042
Efficient Exploration in Constrained Environments with Goal-Oriented Reference Path. Peer-reviewed

Kei Ota, Yoko Sasaki, Devesh K. Jha, Yusuke Yoshiyasu, Asako Kanezaki

CoRR　abs/2003.01641　2020
Visual Object Search by Learning Spatial Context. Peer-reviewed

Raphael Druon, Yusuke Yoshiyasu, Asako Kanezaki, Alassane Watt

IEEE Robotics Autom. Lett.　5　(2)　1279-1286　2020

DOI： 10.1109/LRA.2020.2967677 　
RotationNet for Joint Object Categorization and Unsupervised Pose Estimation from Multi-view Images. International-journal Peer-reviewed

Asako Kanezaki, Yasuyuki Matsushita, Yoshifumi Nishida

IEEE transactions on pattern analysis and machine intelligence　2019/06/14

DOI： 10.1109/TPAMI.2019.2922640 　

More details Close

We propose a Convolutional Neural Network (CNN)-based model "RotationNet," which takes multi-view images of an object as input and jointly estimates its pose and object category. Unlike previous approaches that use known viewpoint labels for training, our method treats the viewpoint labels as latent variables, which are learned in an unsupervised manner during the training using an unaligned object dataset. RotationNet uses only a partial set of multi-view images for inference, and this property makes it useful in practical scenarios where only partial views are available. Moreover, our pose alignment strategy enables one to obtain view-specific feature representations shared across classes, which is important to maintain high accuracy in both object categorization and pose estimation. Effectiveness of RotationNet is demonstrated by its superior performance to the state-of-the-art methods of 3D object classification on 10- and 40-class ModelNet datasets. We also show that RotationNet, even trained without known poses, achieves comparable performance to the state-of-the-art methods on an object pose estimation dataset. Furthermore, our object ranking method based on classification by RotationNet achieved the first prize in two tracks of the 3D Shape Retrieval Contest (SHREC) 2017. Finally, we demonstrate the performance of real-world applications of RotationNet trained with our newly created multi-view image dataset using a moving USB camera.
Deep learning for multimodal data fusion

Asako Kanezaki, Ryohei Kuga, Yusuke Sugano, Yasuyuki Matsushita

Multimodal Scene Understanding: Algorithms, Applications and Deep Learning　9-39　2019/01/01
Publisher: Elsevier
DOI： 10.1016/B978-0-12-817358-9.00008-1 　
Mobile Robot Motion Planning in Crowds of People Using Deep Reinforced Learning Peer-reviewed

SASAKI Yoko, MATSUO Shusuke, KANEZAKI Asako, TAKEMURA Hiroshi

The Proceedings of JSME annual Conference on Robotics and Mechatronics (Robomec)　2019　(0)　2P2-B03　2019
Publisher: The Japan Society of Mechanical Engineers
DOI： 10.1299/jsmermd.2019.2P2-B03 　

eISSN： 2424-3124

More details Close

The paper proposes a motion planning algorithm using deep reinforced learning algorithm; Asynchronous Actor-Critic Agents(A3C)．For mobile robot navigation tasks in crowded situations, existingpath planning based approaches are limited because the surrounding environment is changing dynamically. The correct motion in such a dynamic environment is underspecified, and reinforced learning approach is suitable to generate applicable motion. We propose A3C based motion planning method to acquire mobile robot motion going through crowds of people situation. The proposed method is evaluated in simulated crowds of pedestrians. The experiment section shows the basic performance depending on training parameters and some generated motion examples in the simulator.
A study on markers for object recognition by Deep learning:-Comparison of recognition rates by AlexNet- Peer-reviewed

KANO Ryoya, WADA Kazuyoshi, KANEZAKI Asako, TOMIZAWA Tetsuo, TANIKAWA Tamio

The Proceedings of JSME annual Conference on Robotics and Mechatronics (Robomec)　2019　(0)　2P1-Q01　2019
Publisher: The Japan Society of Mechanical Engineers
DOI： 10.1299/jsmermd.2019.2P1-Q01 　

eISSN： 2424-3124

More details Close

Recently, it has been required to automate the task of displaying products in convenience stores by robots. In this paper, we use a deep learning-based method to recognize pose and category of product. However, when an object is photographed by a camera, the recognition rate of objects with few features in color and shape is low. Therefore, we made a hypothesis that the recognition rate is improved by adding some features to the product. Two types of color markers were developed and those recognition rates were examined by using AlexNet. As a result, the proposed markers were able to improve the category recognition rate and pose recognition rate of the products.
Salient object detection on hyperspectral images using features learned from unsupervised segmentation task. Peer-reviewed

Nevrez Imamoglu, Guanqun Ding, Yuming Fang, Asako Kanezaki, Toru Kouyama, Ryosuke Nakamura

CoRR　abs/1902.10993　2019
A3C Based Motion Learning for an Autonomous Mobile Robot in Crowds Peer-reviewed

Yoko Sasaki, Syusuke Matsuo, Asako Kanezaki, Hiroshi Takemura

2019 IEEE INTERNATIONAL CONFERENCE ON SYSTEMS, MAN AND CYBERNETICS (SMC)　1036-1042　2019

DOI： 10.1109/SMC.2019.8914201 　

ISSN： 1062-922X
Salient Object Detection on Hyperspectral Images Using Features Learned from Unsupervised Segmentation Task. Peer-reviewed

Nevrez Imamoglu, G. Ding, Y. Fang, Asako Kanezaki, Toru Kouyama, R. Nakamura

IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)　2192-2196　2019
Publisher: IEEE
DOI： 10.1109/ICASSP.2019.8682522 　
GOSELO: Goal-Directed Obstacle and Self-Location Map for Robot Navigation Using Reactive Neural Networks Peer-reviewed

Asako Kanezaki, Jirou Nitta, Yoko Sasaki

IEEE ROBOTICS AND AUTOMATION LETTERS　3　(2)　696-703　2018/04

DOI： 10.1109/LRA.2017.2783400 　

ISSN： 2377-3766
An integration of bottom-up and top-down salient cues on RGB-D data: saliency from objectness versus non-objectness Peer-reviewed

Nevrez Imamoglu, Wataru Shimoda, Chi Zhang, Yuming Fang, Asako Kanezaki, Keiji Yanai, Yoshifumi Nishida

SIGNAL IMAGE AND VIDEO PROCESSING　12　(2)　307-314　2018/02

DOI： 10.1007/s11760-017-1159-7 　

ISSN： 1863-1703

eISSN： 1863-1711
Estimation of Number and Locations of Products in Pictures by Semi-Supervised Learning Peer-reviewed

FUJIHASHI Kazuki, KIMURA Masayuki, KANEZAKI Asako, OZAWA Jun

Proceedings of the Annual Conference of JSAI　2018　(0)　4M204-4M204　2018
Publisher: The Japanese Society for Artificial Intelligence
DOI： 10.11517/pjsai.JSAI2018.0_4M204 　

ISSN： 1347-9881

More details Close

We propose a semi-supervised method for estimating the number and locations of products in pictures. Many existing approaches can estimate objects locations in images by supervised learning which needs images annotated with objects locations. On the other hand, our method needs only numbers of objects in images. The experiment shows effectiveness of our method.
RotationNet: Joint Object Categorization and Pose Estimation Using Multiviews from Unsupervised Viewpoints Peer-reviewed

Asako Kanezaki, Yasuyuki Matsushita, Yoshifumi Nishida

2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)　5010-5019　2018

DOI： 10.1109/CVPR.2018.00526 　

ISSN： 1063-6919
An Integration of Bottom-up and Top-Down Salient Cues on RGB-D Data: Saliency from Objectness vs. Non-Objectness. Peer-reviewed

Nevrez Imamoglu, Wataru Shimoda, Chi Zhang, Yuming Fang, Asako Kanezaki, Keiji Yanai, Yoshifumi Nishida

CoRR　abs/1807.01532　2018
UNSUPERVISED IMAGE SEGMENTATION BY BACKPROPAGATION Peer-reviewed

Asako Kanezaki

2018 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP)　1543-1547　2018

DOI： 10.1109/ICASSP.2018.8462533 　
"Change the Changeable" Framework for Implementation Research in Health. Peer-reviewed

Mikiko Oono, Yoshifumi Nishida, Koji Kitamura, Asako Kanezaki, Tatsuhiro Yamanaka

Proceedings of the 10th International Conference on Computer Supported Education, CSEDU 2018, Funchal, Madeira, Portugal, March 15-17, 2018, Volume 2.　361-368　2018
Publisher: SciTePress
DOI： 10.5220/0006691303610368 　
cvpaper.challenge in 2016: Futuristic Computer Vision through 1, 600 Papers Survey. Peer-reviewed

Hirokatsu Kataoka, Soma Shirakabe, Yun He, Shunya Ueta, Teppei Suzuki, Kaori Abe, Asako Kanezaki, Shinichiro Morita, Toshiyuki Yabe, Yoshihiro Kanehara, Hiroya Yatsuyanagi, Shinya Maruyama, Ryousuke Takasawa, Masataka Fuchida, Yudai Miyashita, Kazushige Okayasu, Yuta Matsuzaki

CoRR　abs/1707.06436　2017

More details Close

The paper gives futuristic challenges disscussed in the cvpaper.challenge. In 2015 and 2016, we thoroughly study 1,600+ papers in several conferences/journals such as CVPR/ICCV/ECCV/NIPS/PAMI/IJCV.
Multi-task Learning using Multi-modal Encoder-Decoder Networks with Shared Skip Connections Peer-reviewed

Ryohei Kuga, Asako Kanezaki, Masaki Samejima, Yusuke Sugano, Yasuyuki Matsushita

2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION WORKSHOPS (ICCVW 2017)　2018-　403-411　2017

DOI： 10.1109/ICCVW.2017.54 　

ISSN： 2473-9936
Large-Scale 3D Shape Retrieval from ShapeNet Core55. Peer-reviewed

Manolis Savva, Fisher Yu, Hao Su, Asako Kanezaki, Takahiko Furuya, Ryutarou Ohbuchi, Zhichao Zhou, Rui Yu, Song Bai, Xiang Bai, Masaki Aono, Atsushi Tatsuma, Spyridon Thermos, Apostolos Axenopoulos, Georgios Th. Papadopoulos, Petros Daras, Xiao Deng, Zhouhui Lian, Bo Li 0013, Henry Johan, Yijuan Lu, Sanjeev Mk

Eurographics Workshop on 3D Object Retrieval, EG 3DOR　89-98　2017
Publisher: Eurographics Association
DOI： 10.2312/3dor.20171050 　

ISSN： 1997-0471 1997-0463
RGB-D to CAD Retrieval with ObjectNN Dataset. Peer-reviewed

Binh-Son Hua, Quang-Trung Truong, Minh-Khoi Tran, Quang-Hieu Pham, Asako Kanezaki, Tang Lee, HungYueh Chiang, Winston H. Hsu, Bo Li 0013, Yijuan Lu, Henry Johan, Shoki Tashiro, Masaki Aono, Minh-Triet Tran, Viet-Khoi Pham, Hai-Dang Nguyen, Vinh-Tiep Nguyen, Quang-Thang Tran, Thuyen V. Phan, Bao Truong, Minh N. Do, Anh Duc Duong, Lap-Fai Yu, Duc Thanh Nguyen, Sai-Kit Yeung

Eurographics Workshop on 3D Object Retrieval, 3DOR 2017, Lyon, France, April 23-24, 2017　2017
Publisher: Eurographics Association
DOI： 10.2312/3dor.20171048 　
繋げる人工知能:生活機能レジリエント社会のためのスマートリビングネットを用いた生活知能研究

西田佳史, 北村光司, 佐々木洋子, 金崎朝子, SHI Boxin, 大野美喜子, 楠本欣司

計測自動制御学会システムインテグレーション部門講演会(CD-ROM)　17th　ROMBUNNO.3K4‐5　2016/12/15
Recognizing Activities of Daily Living with a Wrist-mounted Camera Peer-reviewed

Katsunori Ohnishi, Atsushi Kanehira, Asako Kanezaki, Tatsuya Harada

2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)　abs/1511.06783　3103-3111　2016

DOI： 10.1109/CVPR.2016.338 　

ISSN： 1063-6919
RotationNet: Learning Object Classification Using Unsupervised Viewpoint Estimation. Peer-reviewed

Asako Kanezaki

CoRR　abs/1603.06208　2016
IBC127: VIDEO DATASET FOR FINE-GRAINED BIRD CLASSIFICATION Peer-reviewed

Tomoaki Saito, Asako Kanezaki, Tatsuya Harada

2016 IEEE INTERNATIONAL CONFERENCE ON MULTIMEDIA & EXPO (ICME)　1-6　2016

DOI： 10.1109/ICME.2016.7552915 　

ISSN： 1945-7871
Recognizing Activities of Daily Living with a Wrist-Mounted Camera. Peer-reviewed

Katsunori Ohnishi, Atsushi Kanehira, Asako Kanezaki, Tatsuya Harada

2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)　3103-3111　2016

DOI： 10.1109/CVPR.2016.338 　

ISSN： 1063-6919
Learning similarities for rigid and non-rigid object detection Peer-reviewed

Asako Kanezaki, Emanuele Rodolà, Daniel Cremers, Tatsuya Harada

Proceedings - 2014 International Conference on 3D Vision, 3DV 2014　114　(230)　720-727　2015/02/06
Publisher: Institute of Electrical and Electronics Engineers Inc.
DOI： 10.1109/3DV.2014.61 　

ISSN： 0913-5685
三次元情報を活用した物体検出の三手法

金崎朝子

ViEWビジョン技術の実利用ワークショップ講演論文集(CD-ROM)　2015　ROMBUNNO.OS5‐O1　2015
Probabilistic Semi-Canonical Correlation Analysis Peer-reviewed

Chie Kamada, Asako Kanezaki, Tatsuya Harada

MM'15: PROCEEDINGS OF THE 2015 ACM MULTIMEDIA CONFERENCE　1131-1134　2015

DOI： 10.1145/2733373.2806299 　
3D Selective Search for Obtaining Object Candidates Peer-reviewed

Asako Kanezaki, Tatsuya Harada

2015 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS (IROS)　82-87　2015

DOI： 10.1109/IROS.2015.7353358 　

ISSN： 2153-0858
Learning Similarities for Rigid and Non-Rigid Object Detection Peer-reviewed

KANEZAKI Asako, RODOLA Emanuele, CREMERS Daniel, HARADA Tatsuya

Technical report of IEICE. PRMU　114　(230)　13-18　2014/10/09
Publisher: The Institute of Electronics, Information and Communication Engineers
ISSN： 0913-5685

More details Close

We propose an optimization method for estimating parameters in graph-theoretical formulations of the matching problem for object detection. Unlike several methods which optimize parameters for graph matching in a way to promote correct correspondences and to restrict wrong ones, our approach aims at improving performance in the more general task of object detection. In our formulation, similarity functions are adjusted so as to increase the overall similarity among a reference model and the observed target, and at the same time reduce the similarity among reference and "non-target" objects. We evaluate the proposed method in two challenging scenarios, demonstrating substantial improvements in both settings.
RGB‐D画像からの物体検出における対応点集合類似度の学習

金崎朝子, RODOLA Emanuele, 原田達也

日本ロボット学会学術講演会予稿集(CD-ROM)　32nd　ROMBUNNO.3I2-03　2014/09/04
Part-Based Geometric Categorization and Object Reconstruction in Cluttered Table-Top Scenes Peer-reviewed

Zoltan-Csaba Marton, Ferenc Balint-Benczedi, Oscar Martinez Mozos, Nico Blodow, Asako Kanezaki, Lucian Cosmin Goron, Dejan Pangercic, Michael Beetz

JOURNAL OF INTELLIGENT & ROBOTIC SYSTEMS　76　(1)　35-56　2014/09

DOI： 10.1007/s10846-013-0011-8 　

ISSN： 0921-0296

eISSN： 1573-0409
Clothing Retrieval Based on Local Similarity with Multiple Images Peer-reviewed

Masaru Mizuochi, Asako Kanezaki, Tatsuya Harada

PROCEEDINGS OF THE 2014 ACM CONFERENCE ON MULTIMEDIA (MM'14)　1165-1168　2014

DOI： 10.1145/2647868.2655021 　
Automatic Image Synthesis from Keywords Using Scene Context Peer-reviewed

Sho Inaba, Asako Kanezaki, Tatsuya Harada

PROCEEDINGS OF THE 2014 ACM CONFERENCE ON MULTIMEDIA (MM'14)　1149-1152　2014

DOI： 10.1145/2647868.2655009 　
Hard Negative Classes for Multiple Object Detection Peer-reviewed

Asako Kanezaki, Sho Inaba, Yoshitaka Ushiku, Yuya Yamashita, Hiroshi Muraoka, Yasuo Kuniyoshi, Tatsuya Harada

2014 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA)　3066-3073　2014

DOI： 10.1109/ICRA.2014.6907300 　

ISSN： 1050-4729

eISSN： 2577-087X
MIRROR REFLECTION INVARIANT HOG DESCRIPTORS FOR OBJECT DETECTION Peer-reviewed

Asako Kanezaki, Yusuke Mukuta, Tatsuya Harada

2014 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP)　1594-1598　2014

DOI： 10.1109/ICIP.2014.7025319 　

ISSN： 1522-4880
Weakly-supervised multi-class object detection using multi-type 3D features

Asako Kanezaki, Yasuo Kuniyoshi, Tatsuya Harada

MM 2013 - Proceedings of the 2013 ACM Multimedia Conference　605-608　2013

DOI： 10.1145/2502081.2502159 　
Simultaneous training of multi-class object detectors via large scale image dataset : introduction of target specific negative classes Peer-reviewed

KANEZAKI Asako, INABA Sho, USHIKU Yoshitaka, YAMASHITA Yuya, MURAOKA Hiroshi, HARADA Tatsuya, KUNIYOSHI Yasuo

電子情報通信学会技術研究報告 : 信学技報　112　(198)　105-112　2012/09/02
Publisher: The Institute of Electronics, Information and Communication Engineers
ISSN： 0913-5685

More details Close

We propose an efficient method to train multiple object detectors simultaneously using a large-scale image dataset. The one-vs-all approach that optimizes the boundary between positive samples from a target class and negative samples from the others has been the most standard approach for object detection. However, because this approach trains each object detector independently, the likelihoods are not balanced between object classes. The proposed method combines ideas derived from both detection and classification in order to balance the scores across all object classes. We optimized the boundary between target classes and their hard-negative samples, just as in detection, while simultaneously balancing the detector likelihoods across obj ect classes, as done in multi-class classification. We evaluated the performances on multi-class object detection using a subset of the ImageNet Large Scale Visual Recognition Challenge(ILSVRC) 2011 dataset and showed our method outperformed a de facto standard method.
Simultaneous training of multi-class object detectors via large scale image dataset : introduction of target specific negative classes Peer-reviewed

KANEZAKI Asako, INABA Sho, USHIKU Yoshitaka, YAMASHITA Yuya, MURAOKA Hiroshi, HARADA Tatsuya, KUNIYOSHI Yasuo

Technical report of IEICE. PRMU　112　(197)　105-112　2012/09/02
Publisher: The Institute of Electronics, Information and Communication Engineers
ISSN： 0913-5685

More details Close

We propose an efficient method to train multiple object detectors simultaneously using a large-scale image dataset. The one-vs-all approach that optimizes the boundary between positive samples from a target class and negative samples from the others has been the most standard approach for object detection. However, because this approach trains each object detector independently, the likelihoods are not balanced between object classes. The proposed method combines ideas derived from both detection and classification in order to balance the scores across all object classes. We optimized the boundary between target classes and their hard-negative samples, just as in detection, while simultaneously balancing the detector likelihoods across object classes, as done in multi-class classification. We evaluated the performances on multi-class object detection using a subset of the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2011 dataset and showed our method outperformed a de facto standard method.
Simultaneous training of multi-class object detectors via large scale image dataset introduction of target specific negative classes Peer-reviewed

Asako Kanezaki, Sho Inaba, Yoshitaka Ushiku, Yuya Yamashita, Hiroshi Muraoka, Tatsuya Harada, Yasuo Kuniyoshi

IPSJ SIG Notes. CVIM　2012　(17)　1-8　2012/08/26
Publisher: Information Processing Society of Japan (IPSJ)
ISSN： 0919-6072

More details Close

We propose an efficient method to train multiple object detectors simultaneously using a large-scale image dataset.The one-vs-all approach that optimizes the boundary between positive samples from a target class and negative samples from the others has been the most standard approach for object detection. However, because this approach trains each object detector independently, the likelihoods are not balanced between object classes. The proposed method combines ideas derived from both detection and classification in order to balance the scores across all object classes. We optimized the boundary between target classes and their hard-negative samples, just as in detection, while simultaneously balancing the detector likelihoods across object classes, as done in multi-class classification. We evaluated the performances on multi-class object detection using a subset of the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2011 dataset and showed our method outperformed a de facto standard method.
Simultaneous training of multi-class object detectors via large scale image dataset introduction of target specific negative classes

金崎朝子, 稲葉翔, 牛久祥孝, 山下裕也, 村岡宏是, 原田達也, 國吉康夫

電子情報通信学会技術研究報告　112　(197(PRMU2012 30-50))　105-112　2012/08/26
Publisher: 一般社団法人情報処理学会
ISSN： 0913-5685

More details Close

We propose an efficient method to train multiple object detectors simultaneously using a large-scale image dataset.The one-vs-all approach that optimizes the boundary between positive samples from a target class and negative samples from the others has been the most standard approach for object detection. However, because this approach trains each object detector independently, the likelihoods are not balanced between object classes. The proposed method combines ideas derived from both detection and classification in order to balance the scores across all object classes. We optimized the boundary between target classes and their hard-negative samples, just as in detection, while simultaneously balancing the detector likelihoods across object classes, as done in multi-class classification. We evaluated the performances on multi-class object detection using a subset of the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2011 dataset and showed our method outperformed a de facto standard method.
Voxelized Shape and Color Histograms for RGB-D Peer-reviewed

Asako Kanezaki, Zoltan-Csaba Marton, Dejan Pangercic, Tatsuya Harada, Yasuo Kuniyoshi, Michael Beetz

IEEE IROS Workshop on Active Semantic Perception and Object Search in the Real World (ASP-AVS-11)　2011/09
Fast Object Detection for Robots in a Cluttered Indoor Environment Using Integral 3D Feature Table Peer-reviewed

Asako Kanezaki, Takahiro Suzuki, Tatsuya Harada, Yasuo Kuniyoshi

2011 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA)　4026-4033　2011

DOI： 10.1109/ICRA.2011.5980129 　

ISSN： 1050-4729

eISSN： 2577-087X
Scale and Rotation Invariant Color Features for Weakly-Supervised Object Learning in 3D Space Peer-reviewed

Asako Kanezaki, Tatsuya Harada, Yasuo Kuniyoshi

2011 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION WORKSHOPS (ICCV WORKSHOPS)　617-624　2011

DOI： 10.1109/ICCVW.2011.6130300 　
Partial matching of real textured 3D objects using color cubic higher-order local auto-correlation features Peer-reviewed

Asako Kanezaki, Tatsuya Harada, Yasuo Kuniyoshi

VISUAL COMPUTER　26　(10)　1269-1281　2010/10

DOI： 10.1007/s00371-010-0521-3 　

ISSN： 0178-2789

eISSN： 1432-2315
High-speed 3D Object Recognition Using Additive Features in A Linear Subspace Peer-reviewed

Asako Kanezaki, Hideki Nakayama, Tatsuya Harada, Yasuo Kuniyoshi

2010 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA)　3128-3134　2010

DOI： 10.1109/ROBOT.2010.5509271 　

ISSN： 1050-4729

eISSN： 2577-087X
Application of High-speed 3D Object Recognition in the Real World Using Integral Features and Linear Subspace Method Peer-reviewed

KANEZAKI Asako, HARADA Tatsuya, KUNIYOSHI Yasuo

IEICE technical report　109　(306)　207-212　2009/11/19
Publisher: The Institute of Electronics, Information and Communication Engineers
ISSN： 0913-5685

More details Close

As a previous work, we proposed a method of high-speed 3D object recognition using linear subspace method and our 3D features. This method enables matching of 3D models in a database and partial models with any size in any posture in a 3D scene, with highly short computation time. In this paper, we experiment on the performance of our method using real 3D models and 3D scenes, which are measured in a day-to-day human environment. In the experiments we compare our method with conventional methods using Spin-Images and Textured Spin-Images. Moreover, we demonstrate and discuss the performance of our method on searching for objects in a real 3D scene.
The Development of Color CHLAC Features for Object Exploration in 3D Map Peer-reviewed

HARADA Tatsuya, KANEZAKI Asako, KUNIYOSHI Yasuo

Journal of the Robotics Society of Japan　27　(7)　749-758　2009/09/15
Publisher: The Robotics Society of Japan
DOI： 10.7210/jrsj.27.749 　

ISSN： 0289-1824

eISSN： 1884-7145

More details Close

We present a new feature “Color Cubic Higher-order Auto-Correlation (Color CHLAC) features” to recognize objects in the real world versatilely and robustly. The new features satisfy the necessary functions for the exploration of objects in a three-dimensional map. In order to search and retrieve objects in a three-dimensional map, the features should have the co-occurrence of textures and shapes, robustness for partial observations and noise, ability to adapt a widespread environment, scalability, and invariance for many transformations. We studied experiments both in a simulation and a real environment for recognition of objects, which have many kinds of shapes and textures, and then we showed that our proposed features obtain high recognition accuracy in both situations.
Partial matching for real textured 3D objects using color cubic higher-order local auto-correlation features

A. Kanezaki, T. Harada, Y. Kuniyoshi

Eurographics Workshop on 3D Object Retrieval, EG 3DOR　9-12　2009

DOI： 10.2312/3DOR/3DOR09/009-012 　

ISSN： 1997-0463 1997-0471
The development of Color CHLAC features for object exploration in 3D map

金崎朝子, 原田達也, 國吉康夫

日本ロボット学会学術講演会予稿集(CD-ROM)　26th　(7)　ROMBUNNO.1L3-06-758　2008/09/09
Publisher: The Robotics Society of Japan
DOI： 10.7210/jrsj.27.749 　

ISSN： 0289-1824

eISSN： 1884-7145

More details Close

We present a new feature “Color Cubic Higher-order Auto-Correlation (Color CHLAC) features” to recognize objects in the real world versatilely and robustly. The new features satisfy the necessary functions for the exploration of objects in a three-dimensional map. In order to search and retrieve objects in a three-dimensional map, the features should have the co-occurrence of textures and shapes, robustness for partial observations and noise, ability to adapt a widespread environment, scalability, and invariance for many transformations. We studied experiments both in a simulation and a real environment for recognition of objects, which have many kinds of shapes and textures, and then we showed that our proposed features obtain high recognition accuracy in both situations.

Show all ︎Show first 5

Misc. 7

Introduction to Basic and Deep Learning Based Methods for Unsupervised Image Segmentation

KANEZAKI Asako

Medical Imaging Technology　39　(4)　142-147　2021/09/25
Publisher: The Japanese Society of Medical Imaging Technology
DOI： 10.11409/mit.39.142 　

ISSN： 0288-450X

eISSN： 2185-3193

More details Close

Unsupervised image segmentation is an important technique in various research fields such as medical image processing. Basically, unsupervised image segmentation is based on some hand-crafted features and clustering pixels in a way that takes into account feature similarity and spatial continuity. In contrast, the authors proposed a method that applies unsupervised learning of convolutional neural networks (CNNs) to image segmentation. The proposed CNN estimates to which cluster each pixel in the input image belongs, as in a general supervised image segmentation task. However, it does not require any supervisory signals of pixel labels or network pre-training, and the network is trained only after the target image is input. In this paper, we describe the conventional basics of such unsupervised image segmentation, as well as the authorʼs proposed method using deep learning.
5分で分かる!? 有名論文ナナメ読み：Zhang, Qi and Goldman, Sally A : EM-DD : An Improved Multiple-Instance Learning Technique

金崎朝子

情報処理　62　(2)　100-101　2021/01/15
Publisher: [出版社不明]
DOI： 10.20729/00208933 　

More details Close

弱教師あり学習の一種であるMultiple-Instance学習では，インスタンスの集合に対して正か負かの教師ラベルが与えられる．正の集合は少なくとも1つ以上の正のインスタンスを含み，負の集合は負のインスタンスのみを含む．このように定義された「集合のラベル」を手がかりとして正のインスタンスを発見する．Multiple-Instance学習を解くことは，何らかの事象の原因を突き止める行為に近い．たとえば，薬物活性予測問題において，さまざまな分子構成の薬物を集めて実験することで，活性化の原因となる分子の低エネルギー構造体を発見することも可能だ．本稿では，このような問題を効率的に解くことが可能なEM-DDという手法を紹介する．
Robot Motion Planning in Dynamic Environments

Ota Kei, Kanezaki Asako

Journal of the Robotics Society of Japan　39　(7)　581-586　2021
Publisher: The Robotics Society of Japan
DOI： 10.7210/jrsj.39.581 　

ISSN： 0289-1824

eISSN： 1884-7145
A Report on MIRU2016 Young Researchers' Program Peer-reviewed

舩冨卓哉, 石井雅人, 井上中順, 金崎朝子, 高橋康輔, 道満恵介, 吉岡隆宏

電子情報通信学会技術研究報告 = IEICE technical report : 信学技報　116　(412)　283-290　2017/01/19
Publisher: 電子情報通信学会
ISSN： 0913-5685
A Report on MIRU2016 Young Researchers' Program Peer-reviewed

舩冨卓哉, 石井雅人, 井上中順, 金崎朝子, 高橋康輔, 道満恵介, 吉岡隆宏

電子情報通信学会技術研究報告 = IEICE technical report : 信学技報　116　(411)　283-290　2017/01/19
Publisher: 電子情報通信学会
ISSN： 0913-5685
MIRU2016若手プログラム実施概要と次回の企画紹介 Peer-reviewed

舩冨卓哉, 石井雅人, 井上中順, 金崎朝子, 高橋康輔, 道満恵介, 吉岡隆宏, 浦西友樹

情報・システムソサイエティ誌　21　(4)　16-22　2017
Publisher: 一般社団法人電子情報通信学会
DOI： 10.1587/ieiceissjournal.21.4_16 　

ISSN： 2189-9797

eISSN： 2189-9819
部分空間法とカラー立体高次局所自己相関特徴を用いた高速三次元物体認識

金崎朝子

画像の認識理解シンポジウム (MIRU), 2009　103-110　2009

Show all ︎Show first 5

Books and Other Publications 1

コンピュータビジョン ―広がる要素技術と応用― (未来へつなぐデジタルシリーズ 37)

米谷竜, 斎藤英雄, 池畑諭, 牛久祥孝, 内山英昭, 内海ゆづ子, 小野峻佑, 片岡裕雄, 金崎朝子, 川西康友, 齋藤真樹, 櫻田健, 高橋康輔, 松井勇佑, 米谷竜, 斎藤英雄

共立出版　2018/06/28

ISBN: 4320123573

Industrial Property Rights 2

抽出装置、方法及びプログラム

金崎朝子, 伊藤聡

Property Type: Patent
特徴ベクトル算出装置、特徴ベクトル算出方法及びプログラム

原田達也, 金崎朝子, 國吉康夫

Property Type: Patent

Research Projects 7

ロボットによる三次元環境記述に関する研究

金崎朝子

Offer Organization: 日本学術振興会

System: 科学研究費助成事業

Category: 基盤研究(B)

Institution: 東京科学大学

2025/04 - 2030/03
生活空間セマンティクス駆動型ロボットに関する研究

金崎朝子

Offer Organization: 科学技術振興機構

Category: 戦略的な研究開発の推進/創発的研究支援事業

2021 - 2027

More details Close

ユーザにとって有用な情報の収集を行動目的とする生活空間セマンティクス駆動型ロボットを提案します。ロボットは自律的に環境内を移動し、膨大なセンサ情報の中から有用であると判断した情報のみを抽出してデータベースに蓄積します。ユーザフィードバックによる情報有用度の再計算を行い、ロボットの行動則を強化学習により更新します。高度認識技術と強化学習を組み合わせた新しい統合的技術を提案し、これを実現します。
Zeroshot learning of real-world AI by fusing large deep learning models and 3D virtual world

Offer Organization: Japan Society for the Promotion of Science

System: Grants-in-Aid for Scientific Research

Category: Grant-in-Aid for Scientific Research (B)

Institution: National Institute of Advanced Industrial Science and Technology

2023/04/01 - 2026/03/31
微分可能クラスタリングによる教師なし画像セグメンテーションの深層学習に関する研究

金崎朝子

Offer Organization: 日本学術振興会

System: 科学研究費助成事業若手研究

Category: 若手研究

Institution: 東京工業大学

2020/04/01 - 2023/03/31

More details Close

画像セグメンテーションは物体等のまとまり毎に画像領域を分割するタスクであり，画像処理の本質的な課題の一つである．従来手法では色やテクスチャを表す画像特徴量を人間が定義し，特徴量の類似度に基づいて領域を分割する方法が主に使われていた．これに対し，昨今の深層学習を用いた手法によれば，車，人，道路等の意味的なまとまりを持った画像領域を分割しラベルを付与するセマンティックセグメンテーションが可能になってきている．しかしながら，深層学習には大量の教師データが必要である．そこで本研究は，教師データを一切必要としない教師なし深層学習による画像セグメンテーションを開発する．昨年度は，これを実現するための微分可能クラスタリングという基盤技術を提案し，理論を確立するとともに，様々なデータセットで有効性を評価した．提案手法は入力画像に対し，特徴抽出モジュールと画素クラスタリングモジュールをEnd-to-Endに学習することで，双方を最適化する．さらに，二次元画像だけでなく動画像データのセグメンテーションへ応用し，様々なアプリケーションへと発展させた．本研究成果について，今年度は「教師なし画像セグメンテーションのベーシックな手法と深層学習ベースの手法の紹介」という論文タイトルで，日本医用画像工学会(JAMIT)誌「MEDICAL IMAGING TECHNOLOGY 39(4)」の特集論文を寄稿した．さらに，第15回IEEE Signal Processing Society (SPS) Japan Student Journal Paper Awardを受賞した．
Research on automatic generation of drawing songs for easy-to-understand object descriptions

Kanezaki Asako

Offer Organization: Japan Society for the Promotion of Science

System: Grants-in-Aid for Scientific Research Grant-in-Aid for Challenging Exploratory Research

Category: Grant-in-Aid for Challenging Exploratory Research

Institution: National Institute of Advanced Industrial Science and Technology

2016/04/01 - 2019/03/31

More details Close

With the aim of explaining unknown objects in images with easy-to-understand expressions, we tackled to develop elemental technology for the automatic generation of drawing songs based on general object recognition. In order to describe an unknown object, humans generate a metaphorical expression such as "Like xxx" or "Like placing xxx on yyy", using some common and imaginable objects. In order to develop such a system, we proposed a novel 3D object recognition method using multi-view images. We also proposed an unsupervised image segmentation method that decomposes unknown objects in images into separated regions without using any prior knowledge.
Action planning of autonomous robots that take pictures for construction of real-world knowledge database

Kanezaki Asako

Offer Organization: Japan Society for the Promotion of Science

System: Grants-in-Aid for Scientific Research Grant-in-Aid for Research Activity start-up

Category: Grant-in-Aid for Research Activity start-up

Institution: The University of Tokyo

2014/08/29 - 2016/03/31

More details Close

First, we proposed an optimization method for estimating the parameters that typically appear in graph theoretical formulations of the matching problem for object detection. Although several methods have been proposed to optimize parameters for graph matching in a way to promote correct correspondences and to restrict wrong ones, our approach is novel in the sense that it aims at improving performance in the more general task of object detection. We presented this work at 3DV 2014 and also achieved 30th (2015) RSJ Young Investigation Excellence Award. Second, to detect unknown objects in the real world, we proposed a new method for obtaining object candidates in 3D space. Our method requires no learning, has no limitation of object properties such as compactness or symmetry, and therefore produces object candidates using a completely general approach. We presented this work at IROS 2015 and also published open source code.
実物体インターネットの実現に向けた、ロボットによる三次元認知地図の獲得と利用

金崎朝子

Offer Organization: 日本学術振興会

System: 科学研究費助成事業特別研究員奨励費

Category: 特別研究員奨励費

Institution: 東京大学

2010 - 2012

More details Close

博士後期課程の最終年度であったため,これまでに得られた研究成果である知見をまとめ,「対象毎の負例クラスの導入による実世界からの多クラス物体認識」と題する博士論文の執筆に従事した.従来,コンピュータビジョン分野において,画像に写っている物体の名前やカテゴリ名を特定する"物体識別"タスクと,画像から対象物体の位置を同定する"物体検出"タスクが広く研究されてきた.しかしながら,「何がどこに存在するか」を判断するタスク,すなわち"物体識別兼検出"タスクは,その重要性に相反して深く議論されてこなかった.また,これらのタスクは主にインターネット等に存在する画像データセットを対象とした機械学習により実現されてきたが,物体の位置や姿勢変化による見え方の変化,また照明変動や物理的な変形への対処は依然として難解な課題であった.本論文では,多クラス物体認識の問題を定式化し,「物体ラベルに対応した物体領域の発見」と「物体認識器の最適化」の二つの課題の解決を要求機能として述べた上で,多数の物体の候補領域に対して識別と検出を同時に行う手法の提案を行った.また評価実験として,100個の対象物体クラスを含む雑多な実環境計測三次元データからの多クラス物体認識を行った.本実験では,100個中の70個の対象物体について正しい領域の発見に成功し,また後段の物体認識器の最適化処理を行うことで,信頼度の高い出力結果の精度をより向上させることができた.このとき,提案手法によって正解の物体の認識率を向上させ,かつ誤認識率を抑えることが可能であった. 本論文の内容は,特別研究員研究課題「実物体インターネットの実現に向けた、ロボットによる三次元認知地図の獲得と利用」において最も根幹となる「実世界からの多クラス物体認識」技術開発において真に必要不可欠な第一歩を踏み出したものであり,滞りなく研究遂行が行われたといえる.

Show all Show first 5