• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.12
• /
• pp.3383-3397
• /
• 2023

Scene graphs serve as semantic abstractions of images and play a crucial role in enhancing visual comprehension and reasoning. However, the performance of Scene Graph Generation is often compromised when working with biased data in real-world situations. While many existing systems focus on a single stage of learning for both feature extraction and classification, some employ Class-Balancing strategies, such as Re-weighting, Data Resampling, and Transfer Learning from head to tail. In this paper, we propose a novel approach that decouples the feature extraction and classification phases of the scene graph generation process. For feature extraction, we leverage a transformer-based architecture and design an adaptive calibration function specifically for predicate classification. This function enables us to dynamically adjust the classification scores for each predicate category. Additionally, we introduce a Distribution Alignment technique that effectively balances the class distribution after the feature extraction phase reaches a stable state, thereby facilitating the retraining of the classification head. Importantly, our Distribution Alignment strategy is model-independent and does not require additional supervision, making it applicable to a wide range of SGG models. Using the scene graph diagnostic toolkit on Visual Genome and several popular models, we achieved significant improvements over the previous state-of-the-art methods with our model. Compared to the TDE model, our model improved mR@100 by 70.5% for PredCls, by 84.0% for SGCls, and by 97.6% for SGDet tasks.

• The KIPS Transactions:PartB
• /
• v.13B no.5 s.108
• /
• pp.499-506
• /
• 2006

This Paper describes a fast and stable camera pose estimation method for real-time augmented reality systems. From the feature tracking results of a marker on a single frame, we estimate the camera rotation matrix and the translation vector. For the camera pose estimation, we use the shape factorization method based on the scaled orthographic Projection model. In the scaled orthographic factorization method, all feature points of an object are assumed roughly at the same distance from the camera, which means the selected reference point and the object shape affect the accuracy of the estimation. This paper proposes a flexible and stable selection method for the reference point. Based on the proposed method, we implemented a video augmentation system that inserts virtual 3D objects into the input video frames. Experimental results showed that the proposed camera pose estimation method is fast and robust relative to the previous methods and it is applicable to various augmented reality applications.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.18 no.1
• /
• pp.100-108
• /
• 2008

This paper proposes a method of adaptively generating a gait pattern of biped robot. The gait synthesis is based on human's gait pattern analysis. The proposed method can easily be applied to generate the natural and stable gait pattern of any biped robot. To analyze the human's gait pattern, sequential images of the human's gait on the sagittal plane are acquired from which the gait control values are extracted. The gait pattern of biped robot on the sagittal plane is adaptively generated by a genetic algorithm using the human's gait control values. However, galt trajectories of the biped robot on the sagittal Plane are not enough to construct the complete gait pattern because the bided robot moves on 3-dimension space. Therefore, the gait pattern on the frontal plane, generated from Zero Moment Point (ZMP), is added to the gait one acquired on the sagittal plane. Consequently, the natural and stable walking pattern for the biped robot is obtained.

• Korean Institute of Interior Design Journal
• /
• v.14 no.3 s.50
• /
• pp.114-121
• /
• 2005

Louis 1. Kahn was a wise architect who learned from history. He developed his own unique architecture by combining his creative sense with design principles and vocabularies that can be found in historical architecture. When restricting a space, he surrounded the space with thick walls as it had been done in historical buildings. The interior space encompassed by this method became a center-oriented and stable space. The objective of this study is to find the characteristics of Kahn's interior spaces by analyzing his projects in terms of space, form, daylight and materials. For this purpose, five works that are considered to have significance from the aspect of interior design were selected and analyzed. The characteristics realized through this study are as follows. A) Spatial features: 1) Generally speaking, each required space has been arranged symmetrically. 2) Being clearly defined as the main space, the subsidiary space, or the service space, each space also was placed very functionally. 3) The space encompassed by thick walls became a center-oriented, stable space. And in most case, it was characterized as a dark space. B) Formative features: 4) The space was defined as a basic solid such as a cylinder, a hexahedron, and an octagonal box, and was developed into a complex shape by the recessed windows. 5) Historical vocabularies such as an arch, a vault, and a dome were reinterpreted in new ways by kahn's own eyes. 6) Haying diverse shapes, the skylights enrich the space in terms of form. C) Daylight feature: 7) The vertical light entering through the skylights creates a solemn and mysterious atmosphere. 8) Given the shadows from the windows that change according to time, the interior space becomes a very vivid space. D) Material feature: 9) Harmonized with cold and smooth materials such as exposed concrete, metal, and glass, the interior space provides a modern atmosphere. 10) Warm appearing wood was used for furniture and part of walls or floors. The effective use of wood takes on a role that is quite complementary to the cold ambience of the smooth and cold materials. 11) With flexibility In building shapes, the concrete becomes the form-endowing materials.

• Journal of Physiology & Pathology in Korean Medicine
• /
• v.25 no.6
• /
• pp.1108-1112
• /
• 2011

Voice can be widely used to classify the four constitution types and to recognize one's health condition from extracting meaningful features as physical quantity in traditional Korean medicine or Western medicine. In this paper, we proposed the method to update the standard operating procedure (SOP) to acquire and record voices for extracting stable vocal features since they are sensitive to the variation of a subject's utterance. At first, we obtained pitch frequencies from vowels and the sentence and intensity form the sentence as features with voices acquired under subjects' utterance conditions and then the deviation ratios of features from median values according to the utterance conditions were obtained and the condition to minimize the ratio was selected as a new SOP. As a result, we decided the SOP for a subject to utter vowels with the length of 2s~1s and sentences with over 2s interval between them after practice, in consideration of the deviation and qualitative requirements. Stable voice features obtained from updated SOP produce accurate diagnosis, which will be developed and simplified for using in the u-Healthcare system of personalized medicine.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.8 no.1
• /
• pp.31-37
• /
• 2008

In this paper, we propose an entropy-based routing protocol to effectively support both stable route construction and route lifetime in Mobile Ad-hoc Wireless Sensor Networks (MAWSN). The basic idea and feature of the proposed routing protocol are as follows. First, we construct the stable routing routes based on entropy concept using mobility of mobile nodes. Second, we consider a realistic approach, in the points of view of the MAWSN, based on mobile sensor nodes as well as fixed sensor nodes in sensor fields while the conventional research for sensor networks focus on mainly fixed sensor nodes. The performance evaluation of the proposed routing protocol is performed via simulation using OPNET(Optimized Network Engineering Tool) and analysis. The results of the performance evaluation show that the proposed routing protocol can efficiently support both the construction of stable route and route lifetime in mobile ad-hoc wireless networks.

• Korean Journal of Oriental Medicine
• /
• v.17 no.3
• /
• pp.97-104
• /
• 2011

Objectives : In this paper, we proposed a method for selecting reliable variables from various vocal features such as frequency derivative features, frequency band ratios, intensities of 5 vowels and an intensity of a sentence, since some features are sensitive to the variation of a subject's utterance. Methods : To obtain the reliable voice variables, the coefficient of variation (CV) was used as the index to evaluate the level of reliability. Since the distributions of a few features are not Gaussian, but are instead skewed to the right or left, we transformed the features by taking the log or square root. Moreover, the definition of the variables that are suitable to represent the vocal property was explained and analyzed. Results : At first, we recorded the vowels and the sentence five times both in the morning and afternoon of the same day, totally ten recordings from each of six subjects (three males and three females). We then analyzed the CVs of each subject's voice to obtain the stable features with a sufficient repeatability. The features having less than 20% CVs for all six subjects were selected. As a result, 92 stable variables from the 222 features were extracted, which included all the transformed variables. Conclusions : Voice can be widely used to classify the four constitution types and to recognize one's health condition from extracting meaningful features as physical quantity in traditional Korean medicine or Western medicine. Therefore, stable voice variables can be useful in the u-Healthcare system of personalized medicine and for improving diagnostic accuracy.

• Journal of Oral Medicine and Pain
• /
• v.30 no.4
• /
• pp.427-432
• /
• 2005

This study have been compared and analyzed voice parameter under the condition of normal voice and auditory condition(noise and music) for 29 patients of orofacial pain and 31 normal people to investigate voice feature and vocal variation for auditory condition of orofacial pain patient. 1. Compared to normal voice, orofacial pain patient showed lower and unstable voice feature which has low F0 rate and high jitter and shimmer rate. 2. Voice of orofacial pain patient showed more relaxed and stable voice feature with low F0 and shimmer rate in the music condition than noise condition. 3. Normal people's voice has no significant difference between music and noise condition even though it has high F0 rate under the noise condition. As a result, orofacial pain patient showed difference of feature and different response for external auditory condition compared to normal voice. Providing of positive emotional environment such as music could be considered for better outcome of oral facial pain patient's functional disability.

• International journal of advanced smart convergence
• /
• v.8 no.4
• /
• pp.104-112
• /
• 2019

This study proposed a robust detection algorithm. It detects face more stably with respect to changes in light and rotation forthe identification of a face shape. The proposed algorithm uses face shape asinput information in a single camera environment and divides only face area through preprocessing process. However, it is not easy to accurately recognize the face area that is sensitive to lighting changes and has a large degree of freedom, and the error range is large. In this paper, we separated the background and face area using the brightness difference of the two images to increase the recognition rate. The brightness difference between the two images means the difference between the images taken under the bright light and the images taken under the dark light. After separating only the face region, the face shape is recognized by using the self-organization feature map (SOFM) algorithm. SOFM first selects the first top neuron through the learning process. Second, the highest neuron is renewed by competing again between the highest neuron and neighboring neurons through the competition process. Third, the final top neuron is selected by repeating the learning process and the competition process. In addition, the competition will go through a three-step learning process to ensure that the top neurons are updated well among neurons. By using these SOFM neural network algorithms, we intend to implement a stable and robust real-time face shape recognition system in face shape recognition.

• Journal of Korea Game Society
• /
• v.17 no.5
• /
• pp.103-112
• /
• 2017

In recent years, expansions and advances of the Internet of Things (IoTs) in a distributed MMOGs (massively multiplayer online games) architecture have resulted in massive growth of data in terms of server workloads. We propose a combing Sparse Autoencoder and one of platforms in MMOGs, ProGReGA. In the process of Sparse Autoencoder, data representation with respect to enhancing the feature is excluded from this set of data. In the process of load balance, the graceful degradation of ProGReGA can exploit the most relevant and less redundant feature of the data representation. We find out that the proposed algorithm have become more stable.