ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper Seeing is Smelling: Localizing Odor-Related Objects in Images
Cited 4 time in scopus Download 9 time Share share facebook twitter linkedin kakaostory
Authors
Sangyun Kim, Junseok Park, Junseong Bang, Haeryong Lee
Issue Date
2018-02
Citation
Augmented Human International Conference (AH) 2018, pp.1-9
Language
English
Type
Conference Paper
DOI
https://dx.doi.org/10.1145/3174910.3174922
Project Code
17HS3100, Olfactory Bio Data based Emotion Enhancement Interactive Content Technology Development, Lee Hae Ryong
Abstract
Research on cross-modal associations between vision and olfaction has shown that odor perception can be strongly influenced by vision including odor-evoked objects. Recently, convolutional neural networks (CNNs), which is mainly applied to image recognition by learning image features, is driving the development of a visual recognition method with promising results. However, there is no attempts to recognition and localization of odor-related objects which allow us to facilitate detection of objects on scene when a congruent odor stimuli is released from scent device. The existing object localization methods require bounding-box annotation indicating the presence of object and location information in an image which is too costly because it is a time-consuming process. Image-level annotation, which indicates the presence of objects in an image, is easier to obtain than bounding-box annotation. In this work, we perform weakly supervised object localization using features of CNNs with only image-level annotation data to detect odor-related objects. We propose a method to classify and localize odor-related objects, and describe its implementation in detail. The experimental results indicate that its performance is comparable to previous CNNs for the classification of odor-related objects.
KSP Keywords
Annotation data, Bounding Box, Convolution neural network(CNN), Image feature, Localization method, Location information(GPS), Recognition method, Weakly supervised, cross-modal, image recognition, object localization