• 한국정보시스템학회지:정보시스템연구
• /
• 제31권4호
• /
• pp.145-160
• /
• 2022

Purpose The purpose of this study is to develop a virtual try-on deep learning model that can efficiently learn front and back clothes images. It is expected that the application of virtual try-on clothing service in the fashion and textile industry field will be vitalization. Design/methodology/approach The data used in this study used 232,355 clothes and product images. The image data input to the model is divided into 5 categories: original clothing image and wearer image, clothing segmentation, wearer's body Densepose heatmap, wearer's clothing-agnosting. We advanced the HR-VITON model in the way of Mixed-Precison, Gradient Accumulation, and sharing model weights. Findings As a result of this study, we demonstrated that the weight-shared MP-GA HR-VITON model can efficiently learn front and back fashion images. As a result, this proposed model quantitatively improves the quality of the generated image compared to the existing technique, and natural fitting is possible in both front and back images. SSIM was 0.8385 and 0.9204 in CP-VTON and the proposed model, LPIPS 0.2133 and 0.0642, FID 74.5421 and 11.8463, and KID 0.064 and 0.006. Using the deep learning model of this study, it is possible to naturally fit one color clothes, but when there are complex pictures and logos as shown in <Figure 6>, an unnatural pattern occurred in the generated image. If it is advanced based on the transformer, this problem may also be improved.

• 한국정보처리학회논문지
• /
• 제3권3호
• /
• pp.449-459
• /
• 1996

본 논문의 퍼지 수리 형태학의 새로운 정의와 신경망을 이용한 이의 구현을 소개 함에 주 목적을 두고 있다. 이 새로운 정의에는 generalized-mean연산자가 중요한 역할을 하고 있다. 본 정의는 신경망을 이용한 구현에 매우 적합할. 연결자 공유 (shared-weight) 신경망의 전반부는 수리 형태적 연산을 수행하기에 적합한 구조를 가 지고 있다. 이 연결자 공유 신경망은 퍼지 수리형태학적 연산을 이용하여 추출 된 특성 정보를 근거로 하여 형태 분류를 수행한다. 따라서, 본 퍼지 정의의 파라 미터들은 신경망의 학습기법을 이용하여 최적화를 기할수 있다. 구조소들(structuring gelements), membership의 값, 그리고 가중 요소(weighting factor)들을 결정하기 위한 학습방법 (learning rule)들이 자세히 열거되어 있다. 적용 예로서 필기체 숫자 인식 문제에 응용한 결과, 퍼지수리 형태학을 이용한 신경망은 이 문제에 있어 현존하는 최고의 결과들과 충분히 견줄만한 결과를 보여주고 있다.

• Journal of Community Nutrition
• /
• 제3권1호
• /
• pp.36-39
• /
• 2001

We developed a junior high school-level computer assisted instruction CAI lesson on obesity and weight control and compared its effectiveness with the lecture/discussion method. we collected pretest and posttest data on knowledge an dinterest in obesity and weight control form 450 students in junior high school We randomly assigned three groups the CAI lesson group a comparable lecture/discussion lesson group and control group There were significant difference among these three groups in posttest knowledge scores and in interest in weight control CAI student's responses to learning by computer were significantly positive.

• 한국음향학회지
• /
• 제21권5호
• /
• pp.430-437
• /
• 2002

다층신경망 (MLP: multilayer perceptron)은 다른 패턴인식 방법에 비해 여러 가지 훌륭한 특성을 가지고 있어 음성인식 및 화자인식 영역에서 폭넓게 사용되고 있다. 그러나 다층신경망의 학습에 일반적으로 사용되는 오류역전파 (EBP: error backpropagation) 알고리즘은 학습시간이 비교적 오래 걸린다는 단점이 있으며, 이는 화자인식이나 화자적응과 같이 실시간 처리를 요구하는 응용에서 상당한 제약으로 작용한다. 패턴인식에 사용되는 학습데이터는 풍부한 중복특성을 내포하고 있으므로 패턴마다 다층신경망의 내부변수를 갱신하는 온라인 계열의 학습방식이 속도의 향상에 상당한 효과가 있다. 일반적인 온라인 오류역전파 알고리즘에서는 가중치 갱신 시 고정된 학습률을 적용한다. 고정 학습률을 적절히 선택함으로써 패턴인식 응용에서 상당한 속도개선을 얻을 수 있지만, 학습률이 고정된 상태에서는 학습이 진행됨에 따라 학습에 기여하는 패턴영역이 달라지는 현상에 효과적으로 대응하지 못하는 문제가 있다. 이 문제에 대해 본 논문에서는 패턴의 기여도에 따라 가변 하는 학습률과 학습에 기여하는 패턴만을 학습에 반영하는 패턴별 가변 학습률 및 학습생략 (COIL: Changing rate and Omitting patterns in Instant Learning)방법을 제안한다. 제안한 COIL의 성능을 입증하기 위해 화자증명과 음성인식을 실험하고 그 결과를 제시한다.

• 예술인문사회 융합 멀티미디어 논문지
• /
• 제8권3호
• /
• pp.57-67
• /
• 2018

앙상블 기법은 기계학습에서 다수의 알고리즘을 사용하여 더 좋은 성능을 내기 위해 사용하는 방법이다. 본 논문에서는 앙상블 기법에서 많이 사용되는 부스팅과 배깅에 대해 소개를 하고, 서포트벡터 회귀, 방사기저함수 네트워크, 가우시안 프로세스, 다층 퍼셉트론을 이용하여 설계한다. 추가적으로 순환신경망과 MOHID 수치모델을 추가하여 실험을 진행한다. 실험적 검증를 위해 사용하는 뜰개 데이터는 7 개의 지역에서 관측된 683 개의 관측 자료다. 뜰개 관측 자료를 이용하여 6 개의 알고리즘과의 비교를 통해 앙상블 기법의 성능을 검증한다. 검증 방법으로는 평균절대오차를 사용한다. 실험 방법은 배깅, 부스팅, 기계학습을 이용한 앙상블 모델을 이용하여 진행한다. 각 앙상블 모델마다 동일한 가중치를 부여한 방법, 차등한 가중치를 부여한 방법을 이용하여 오류율을 계산한다. 가장 좋은 오류율을 나타낸 방법은 기계학습을 이용한 앙상블 모델로서 6 개의 기계학습의 평균에 비해 61.7%가 개선된 결과를 보였다.

• 전자공학회논문지C
• /
• 제34C권7호
• /
• pp.82-91
• /
• 1997

In this paper, we analyze the dynamics of langevine competitive learning neural network based on its fokker-plank equation. From the viewpont of the stochastic differential equation (SDE), langevine competitive learning equation is one of langevine stochastic differential equation and has the diffusin equation on the topological space (.ohm., F, P) with probability measure. We derive the fokker-plank equation from the proposed algorithm and prove by introducing a infinitestimal operator for markov semigroups, that the weight vector in the particular simplex can converge to the globally optimal point under the condition of some convex or pseudo-convex performance measure function. Experimental resutls for pattern recognition of the remote sensing data indicate the superiority of langevine competitive learning neural network in comparison to the conventional competitive learning neural network.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 1999년도 하계종합학술대회 논문집
• /
• pp.1017-1020
• /
• 1999

Recently, a new learning algorithm for multilayer neural networks has been proposed 〔1〕. In the new learning algorithm, each output neuron is considered as a function of weights and the weights are adjusted so that the output neurons produce desired outputs. And the adjustment is accomplished by taking gradients. However, the gradient computation was performed numerically, resulting in a long computation time. In this paper, we derive the all necessary equations so that the gradient computation is performed analytically, resulting in a much faster learning time comparable to the backpropagation. Since the weight adjustments are accomplished by summing the gradients of the output neurons, we will call the new learning algorithm “multi-gradient.” Experiments show that the multi-gradient consistently outperforms the backpropagation.

• 전자공학회논문지B
• /
• 제31B권9호
• /
• pp.133-141
• /
• 1994

Conventional competitive learning algorithms compute the Euclidien distance to determine the winner neuron out of all predetermined output neurons. In such cases, there is a drawback that the performence of the learning algorithm depends on the initial reference(=weight) vectors. In this paper, we propose a new competitive learning algorithm that dynamically generates output neurons. The proposed method generates output neurons by dynamically changing the class thresholds for all output neurons. We compute the similarity between the input vector and the reference vector of each output neuron generated. If the two are similar, the reference vector is adjusted to make it still more like the input vector. Otherwise, the input vector is designated as the reference vector of a new outputneuron. Since the reference vectors of output neurons are dynamically assigned according to input pattern distribution, the proposed method gets around the phenomenon that learning is early determined due to redundant output neurons. Experiments using speech data have shown the proposed method to be superior to existint methods.

• 디지털산업정보학회논문지
• /
• 제17권4호
• /
• pp.85-94
• /
• 2021

The LPR system's trigger sensor makes problem occasionally due to the heave weight of vehicle or the obsolescence equipment. If we replace the hardware sensor to the deep-learning based software sensor in order to generate the trigger signal, LPR system maintenance would be a lot easier. In this paper we proposed the deep-learning sliding window based object detection and tracking algorithm for the LPR system's trigger signal generation. The gate passing vehicle's license plate recognition results are combined into the normal tracking algorithm to catch the position of the vehicle on the trigger line. The experimental results show that the deep learning sliding window based trigger signal generating performance was 100% for the gate passing vehicles including the 5.5% trigger signal position errors due to the minimum bounding box location errors in the vehicle detection process.

• Journal of Semiconductor Engineering
• /
• 제2권1호
• /
• pp.125-129
• /
• 2021

As the interest of deep learning, techniques to control the color of images in image processing field are evolving together. However, there is no clear standard for color, and it is not easy to find a way to represent only the color itself like the color-palette. In this paper, we propose a novel color palette extraction system by chroma fine-tuning with reinforcement learning. It helps to recognize the color combination to represent an input image. First, we use RGBY images to create feature maps by transferring the backbone network with well-trained model-weight which is verified at super resolution convolutional neural networks. Second, feature maps are trained to 3 fully connected layers for the color-palette generation with a generative adversarial network (GAN). Third, we use the reinforcement learning method which only changes chroma information of the GAN-output by slightly moving each Y component of YCbCr color gamut of pixel values up and down. The proposed method outperforms existing color palette extraction methods as given the accuracy of 0.9140.