• IEMEK Journal of Embedded Systems and Applications
• /
• v.16 no.4
• /
• pp.145-151
• /
• 2021

In this paper, we propose a novel visualization technique to explain the predictions of deep neural networks. We use knowledge distillation (KD) to identify the interior of a black-box model for which we know only inputs and outputs. The information of the black box model will be transferred to a white box model that we aim to create through the KD. The white box model will learn the representation of the black-box model. Second, the white-box model generates attention maps for each of its layers using Grad-CAM. Then we combine the attention maps of different layers using the pixel-wise summation to generate a final saliency map that contains information from all layers of the model. The experiments show that the proposed technique found important layers and explained which part of the input is important. Saliency maps generated by the proposed technique performed better than those of Grad-CAM in deletion game.

• ETRI Journal
• /
• v.29 no.3
• /
• pp.353-362
• /
• 2007

In this paper, we propose a novel unsupervised video object extraction algorithm for individual images or image sequences with low depth of field (DOF). Low DOF is a popular photographic technique which enables the representation of the photographer's intention by giving a clear focus only on an object of interest (OOI). We first describe a fast and efficient scheme for extracting OOIs from individual low-DOF images and then extend it to deal with image sequences with low DOF in the next part. The basic algorithm unfolds into three modules. In the first module, a higher-order statistics map, which represents the spatial distribution of the high-frequency components, is obtained from an input low-DOF image. The second module locates the block-based OOI for further processing. Using the block-based OOI, the final OOI is obtained with pixel-level accuracy. We also present an algorithm to extend the extraction scheme to image sequences with low DOF. The proposed system does not require any user assistance to determine the initial OOI. This is possible due to the use of low-DOF images. The experimental results indicate that the proposed algorithm can serve as an effective tool for applications, such as 2D to 3D and photo-realistic video scene generation.

• Proceedings of the KSRS Conference
• /
• 2003.11a
• /
• pp.795-797
• /
• 2003

GIS can supply large quantitiy of useful information about spacial relationship, consequently there exists great potential of GIS support ing for decision making strategies related to urban design and environmental planning. In different discipline areas there exist different information representation methods and considerations having direct or indirect relationship with spacial information, how to use GIS as a tool for facilitating the work in other discipline field, there are two basic questions crucial to the success of these interdiscipline applications. The first one is interoperatability among GIS and other applications, the second is the paradigm difference between GIS and other domain field regarding problem solving. In this paper, we investigate the spatial information of the urban environment provided in the analyses of the urban visual sustain ability, urban daylighting environment and urban wind environment. We also discuss the challenges and opportunities for cross-disciplinary GIS application regarding the aspect of Information Collection, Information Generation, Information Analysis and Information Visualization.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.101.1-101
• /
• 2001

This paper describes a grid mapping for a vision based mobile robot in uncertain indoor environment. The map building is a prerequisite for navigation of a mobile robot and the problem of feature correspondence across two images is well known to be of crucial Importance for vision-based mapping We use a stereo matching algorithm obtained by singular value decomposition of an appropriate correspondence strength matrix. This new correspondence strength means a correlation weight for some local measurements to quantify similarity between features. The visual range data from the reconstructed disparity image form an occupancy grid representation. The occupancy map is a grid-based map in which each cell has some value indicating the probability at that location ...

• 제어로봇시스템학회:학술대회논문집
• /
• 1997.10a
• /
• pp.716-719
• /
• 1997

An integrated program of driving simulator has been developed for excavators using the Motif, OpenGL, and C compiler. The developed program not only offers a GUI but also covers graphic algorithms, therefore, the user can easily run the driving simulator whose components include a simplified visual graphics system. Several graphics technique are combined and applied to the simulator program in order to increase the speed of graphical representation, which access computer memories, mix 2D models with 3D ones, and use the basic position detection method. A text format environment file has been utilized for organizing more flexible driving circumstances.

• School Mathematics
• /
• v.4 no.3
• /
• pp.347-360
• /
• 2002

This study analyzed the mathematical and didactical contexts of the Euclid's proof about Pythagorean theorem and compared with the teaching methods about Pythagorean theorem in school mathematics. Euclid's proof about Pythagorean theorem which does not use the algebraic methods provide students with the spatial intuition and the geometric thinking in school mathematics. Furthermore, it relates to various mathematical concepts including the cosine rule, the rotation, and the transfor-mation which preserve the area, and so forth. Visual demonstrations can help students analyze and explain mathematical relationship. Compared with Euclid's proof, Algebraic proof about Pythagorean theorem is very simple and it supplies the typical example which can give the relationship between algebraic and geometric representation. However since it does not include various spatial contexts, it forbid many students to understand Pythagorean theorem intuitively. Since both approaches have positive and negative aspects, reciprocal complementary role is required in pedagogical aspects.

• Journal of the Korean Institute of Landscape Architecture
• /
• v.45 no.3
• /
• pp.41-53
• /
• 2017

This study was performed to determine what the plant material were selected to reflect in the 2016 Korea Garden Show designer's garden. It was analyzed that plant material was used to display the theme of the garden and to create a specific space. Under the given theme of 'K-Garden, Shinhallyu Garden(new style garden culture) with the most Korean taste', the plants were used to highlight the theme in two types: 'representation' and 'expression'. There were two 'representation' gardens that imitate a particular space of Korean taste and four 'expression' gardens that showcase the designer's thoughts with abstract concepts and concrete objects. Three gardens included both types of garden. The way of revealing the subject with plants was used more for 'expression' than for 'representation'. There were eleven spaces for 'representation' of the Korean taste, a vegetable garden, faucet, pond, field, nature, a Hanok court garden, groves of bamboo, tile roof, stone wall, rock and backyard of a Hanok connected to the mountain. The planting material was used in two ways: reflecting only the ecological characteristics of the plant, and considering the ecological and visual characteristics together. Vegetation plantings reflecting the ecological characteristics were observed in all eleven spaces. Nine of the spaces reflected the growth environment of the plants, but the other two did not reflect the ecological characteristics of the plants, unlike the designer's intention. In the case of the four spaces that considered the ecological and visual characteristics together, color and size were considered visual characteristics. The plants in the seven spaces that included 'expression' as the theme were selected to reflect the visual characteristics in the order of color, shape, texture, and size, rather than reflecting ecological characteristics. A group planting method was applied. When the plants were used as materials for creating space, Norman(1989) analyzed three enclosure factors(overhead, vertical, ground plane). Only two deciduous trees were used in the overhead plane while five species of evergreen shrubs and thirty species of various deciduous plants were used in the vertical plane. There were forty-five species (nine trees and thirty-six herbaceous plants) forming the ground plane, and various herbaceous plants were utilized without duplication in each garden. The designer's garden of the Garden Show played a role in introducing new groundcover plants to the public. Three of the nine gardens did not include ornamental plants, and the use of decorative plants in other gardens was few compared to the number of plantings. In the Korea Garden Show designer's garden, most of the plants were being used with the intention of exposing the theme or architectural uses. In the 2016 Korea Garden Show designer's garden, many species of plants were used as materials for showcasing themes rather than for creating spaces. Also, the method of 'expression' was used more than the method of 'representation' in order to highlight the theme. This indicates that the planting materials reflect visual characteristics such as color, shape, texture, size rather than ecological characteristics.

• KIPS Transactions on Software and Data Engineering
• /
• v.3 no.7
• /
• pp.255-262
• /
• 2014

Our previous research[1] suggested object extraction and modeling method based on Fillmore's case grammar. This approach had not considered of use case extraction and method. To solve this problem, we adopt Fillmore's semantic method as linguistic approach into requirement engineering, which refine fillmore's case grammar for extracting and modeling use cases from customer requirements. This Refined mechanism includes the definition of a structured procedure and the representation of visual notations for 'case' modeling. This paper also proposes the use case decision matrix to identify use case size from extracted use cases based on goal oriented requirement engineering(GoRE), which related with the complexity of use case, and also prioritizes the use cases with this matrix. It demonstrates our proposal with the bank ATM system.

• Journal of the Korea Furniture Society
• /
• v.23 no.3
• /
• pp.252-260
• /
• 2012

Cultural products, applying the design elements of cultural heritage Bekuje content of the product, we chose the useful material that can be used on a desk stand in the smart-phone Sticky Notes (Post-it) for use in modern life. The goal of the development design, developed in the design and excavation element modeling can be taking advantage of the characteristics of traditional decor come connected by a pattern of Baekje, eggplant in harmony with modern life, a new sense was harmony, design. He stressed the beauty of decorative molding to apply discriminatory morphological changes associated with each, a unique pattern of human pattern Jinmyosu Bekuje was elected to the motif of cultural property. You can also use a visual representation of the beauty obviously sophisticated pattern that appears frequently in Baekje, was also emphasized inheritance and traditions of Korea. We also attempted a new way to make sure we can diversify and whether it is possible as a new form, enhance the value of products, from concept to molt general existing tourism products.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2016.04a
• /
• pp.633-635
• /
• 2016

Convolutional Networks (ConvNets) are powerful models that learn hierarchies of visual features, which could also be used to obtain image representations for transfer learning. The basic pipeline for transfer learning is to first train a ConvNet on a large dataset (source task) and then use feed-forward units activation of the trained ConvNet as image representation for smaller datasets (target task). Our key contribution is to demonstrate superior performance of multiple ConvNet layer features over single ConvNet layer features. Combining multiple ConvNet layer features will result in more complex feature space with some features being repetitive. This requires some form of feature selection. We use AdaBoost with single stumps to implicitly select only distinct features that are useful towards classification from concatenated ConvNet features. Experimental results show that using multiple ConvNet layer activation features instead of single ConvNet layer features consistently will produce superior performance. Improvements becomes significant as we increase the distance between source task and the target task.