The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
ZSL(Zero-Shot Learning)은 시각적 특징과 의미적 특징 간의 관계를 학습하여 보이지 않는 클래스의 이미지를 분류하는 것을 목표로 합니다. 기존 연구에서는 다양한 접근 방식을 통해 인식 정확도를 향상시켜 왔지만, 반복적인 최적화가 필요한 계산 집약적인 알고리즘을 사용하고 있습니다. 이 작업에서 우리는 SimpleZSL이라는 매우 간단하고 빠른 방법으로 ZSL 작업을 해결하기 위해 패턴 인식, 즉 가장 가까운 이웃 분류기의 기본 접근 방식을 다시 살펴봅니다. 우리의 알고리즘은 다음 세 가지 간단한 기술로 구성됩니다. (1) 보이는 클래스의 시각적 프로토타입을 얻기 위해 특징 벡터를 평균화하는 것, (2) 보이지 않는 클래스의 시각적 특징을 생성하기 위해 특이값 분해를 통해 의사 역행렬을 계산하는 것, (3) 특징 벡터 사이의 거리를 측정하기 위해 코사인 유사성을 사용하는 최근접 이웃 분류기에 의해 보이지 않는 클래스를 추론합니다. 제안된 방법은 공통 데이터 세트에 대한 실험을 통해 매우 적은 계산 비용으로 좋은 인식 정확도를 달성합니다. 단일 CPU에서 제안된 방법의 실행 시간은 기존 방법의 GPU 구현에 비해 비슷한 정확도로 100배 이상 빠릅니다.
Masayuki HIROMOTO
Fujitsu Limited
Hisanao AKIMA
Fujitsu Limited
Teruo ISHIHARA
Fujitsu Limited
Takuji YAMAMOTO
Fujitsu Limited
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
부
Masayuki HIROMOTO, Hisanao AKIMA, Teruo ISHIHARA, Takuji YAMAMOTO, "SimpleZSL: Extremely Simple and Fast Zero-Shot Learning with Nearest Neighbor Classifiers" in IEICE TRANSACTIONS on Information,
vol. E105-D, no. 2, pp. 396-405, February 2022, doi: 10.1587/transinf.2021EDP7089.
Abstract: Zero-shot learning (ZSL) aims to classify images of unseen classes by learning relationship between visual and semantic features. Existing works have been improving recognition accuracy from various approaches, but they employ computationally intensive algorithms that require iterative optimization. In this work, we revisit the primary approach of the pattern recognition, ı.e., nearest neighbor classifiers, to solve the ZSL task by an extremely simple and fast way, called SimpleZSL. Our algorithm consists of the following three simple techniques: (1) just averaging feature vectors to obtain visual prototypes of seen classes, (2) calculating a pseudo-inverse matrix via singular value decomposition to generate visual features of unseen classes, and (3) inferring unseen classes by a nearest neighbor classifier in which cosine similarity is used to measure distance between feature vectors. Through the experiments on common datasets, the proposed method achieves good recognition accuracy with drastically small computational costs. The execution time of the proposed method on a single CPU is more than 100 times faster than those of the GPU implementations of the existing methods with comparable accuracies.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2021EDP7089/_p
부
@ARTICLE{e105-d_2_396,
author={Masayuki HIROMOTO, Hisanao AKIMA, Teruo ISHIHARA, Takuji YAMAMOTO, },
journal={IEICE TRANSACTIONS on Information},
title={SimpleZSL: Extremely Simple and Fast Zero-Shot Learning with Nearest Neighbor Classifiers},
year={2022},
volume={E105-D},
number={2},
pages={396-405},
abstract={Zero-shot learning (ZSL) aims to classify images of unseen classes by learning relationship between visual and semantic features. Existing works have been improving recognition accuracy from various approaches, but they employ computationally intensive algorithms that require iterative optimization. In this work, we revisit the primary approach of the pattern recognition, ı.e., nearest neighbor classifiers, to solve the ZSL task by an extremely simple and fast way, called SimpleZSL. Our algorithm consists of the following three simple techniques: (1) just averaging feature vectors to obtain visual prototypes of seen classes, (2) calculating a pseudo-inverse matrix via singular value decomposition to generate visual features of unseen classes, and (3) inferring unseen classes by a nearest neighbor classifier in which cosine similarity is used to measure distance between feature vectors. Through the experiments on common datasets, the proposed method achieves good recognition accuracy with drastically small computational costs. The execution time of the proposed method on a single CPU is more than 100 times faster than those of the GPU implementations of the existing methods with comparable accuracies.},
keywords={},
doi={10.1587/transinf.2021EDP7089},
ISSN={1745-1361},
month={February},}
부
TY - JOUR
TI - SimpleZSL: Extremely Simple and Fast Zero-Shot Learning with Nearest Neighbor Classifiers
T2 - IEICE TRANSACTIONS on Information
SP - 396
EP - 405
AU - Masayuki HIROMOTO
AU - Hisanao AKIMA
AU - Teruo ISHIHARA
AU - Takuji YAMAMOTO
PY - 2022
DO - 10.1587/transinf.2021EDP7089
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E105-D
IS - 2
JA - IEICE TRANSACTIONS on Information
Y1 - February 2022
AB - Zero-shot learning (ZSL) aims to classify images of unseen classes by learning relationship between visual and semantic features. Existing works have been improving recognition accuracy from various approaches, but they employ computationally intensive algorithms that require iterative optimization. In this work, we revisit the primary approach of the pattern recognition, ı.e., nearest neighbor classifiers, to solve the ZSL task by an extremely simple and fast way, called SimpleZSL. Our algorithm consists of the following three simple techniques: (1) just averaging feature vectors to obtain visual prototypes of seen classes, (2) calculating a pseudo-inverse matrix via singular value decomposition to generate visual features of unseen classes, and (3) inferring unseen classes by a nearest neighbor classifier in which cosine similarity is used to measure distance between feature vectors. Through the experiments on common datasets, the proposed method achieves good recognition accuracy with drastically small computational costs. The execution time of the proposed method on a single CPU is more than 100 times faster than those of the GPU implementations of the existing methods with comparable accuracies.
ER -