http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Bounding Box CutMix와 표준화 거리 기반의 IoU를 통한 재활용품 탐지
이해진,정희철,Lee, Haejin,Jung, Heechul 대한임베디드공학회 2022 대한임베디드공학회논문지 Vol.17 No.5
In this paper, we developed a deep learning-based recyclable object detection model. The model is developed based on YOLOv5 that is a one-stage detector. The deep learning model detects and classifies the recyclable object into 7 categories: paper, carton, can, glass, pet, plastic, and vinyl. We propose two methods for recyclable object detection models to solve problems during training. Bounding Box CutMix solved the no-objects training images problem of Mosaic, a data augmentation used in YOLOv5. Standardized Distance-based IoU replaced DIoU using a normalization factor that is not affected by the center point distance of the bounding boxes. The recyclable object detection model showed a final mAP performance of 0.91978 with Bounding Box CutMix and 0.91149 with Standardized Distance-based IoU.
Hand-crafted 특징 및 머신 러닝 기반의 은하 이미지 분류 기법 개발
오윤주,정희철,Oh, Yoonju,Jung, Heechul 대한임베디드공학회 2021 대한임베디드공학회논문지 Vol.16 No.1
In this paper, we develop a galaxy image classification method based on hand-crafted features and machine learning techniques. Additionally, we provide an empirical analysis to reveal which combination of the techniques is effective for galaxy image classification. To achieve this, we developed a framework which consists of four modules such as preprocessing, feature extraction, feature post-processing, and classification. Finally, we found that the best technique for galaxy image classification is a method to use a median filter, ORB vector features and a voting classifier based on RBF SVM, random forest and logistic regression. The final method is efficient so we believe that it is applicable to embedded environments.
Diffusion 및 BLIP 모델 기반 이미지 데이터 증강 기법
정탁(Tak Jung),정희철(Heechul Jung) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.6
The importance of data in the field of deep learning is well recognized, particularly in the domain of image classification where a large amount of diverse data is crucial. However, gathering this data can be labor-intensive and costly. In this paper, we propose an algorithm that can augment images and use them for improving image classification performance of convolution neural networks. The proposed algorithm can contribute to solving the data shortage problem of deep learning by generating synthetic data originated from original image. Through empirical evaluations, we validate that our synthetic data not only supplements the original dataset but also enhances the models generalization capabilities. Through this paper, researchers in the field of Computer Vision are expected to develop models of better performance by acquiring image data in a more efficient way than conventional methodologies.
영상기반의 End-to-end 자율주행 알고리즘 개발을 위한 데이터셋 및 평가환경 구축
권순(Soon Kwon),박재형(Jaehyeong Park),정희철(Heechul Jung),정지훈(Jihun Jung),최민국(Min-Kook Choi),Iman R. T.(Iman R. Tayibnapis),이진희(Jin-Hee Lee),원웅재(Woong-Jae Won),김광회(Kwang-Hoe Kim),윤성훈(Sung-Hoon Youn),김태훈(Tae 대한전자공학회 2018 대한전자공학회 학술대회 Vol.2018 No.6
In this paper, we constructed a public dataset for training and evaluation of an algorithm model for Vision based Autonomous Steering Control(V-ASC), and built a benchmark environment to analyze and provide qualitative and quantitative evaluation results. We also developed a baseline V-ASC model based on the handcrafted feature and the newly proposed convolutional neural network (CNN) based end-to-end driving model to verify the evaluation environment of the constructed dataset and simulator. Through the comparative evaluation between the models, we confirmed that the proposed evaluation framework is effective for performance analysis of V-ASC.
딥러닝 기반 벵골어 수기 문자 인식 : Kaggle Bengali.AI 대회를 중심으로
이채현(Chaehyeon Lee),최재협(Jaehyeop Choi),정희철(Heechul Jung) 대한전자공학회 2020 전자공학회논문지 Vol.57 No.9
최근 Kaggle에서는 벵골어 인식을 위한 새로운 데이터 셋이 공개되었고 Bengali.AI라는 국제적인 대회가 열렸다. 본 연구에서는 Bengali.AI 대회에 참가하여 달성한 결과에 대해 공유하고자 한다. 벵골어 문자는 다른 언어에 비해 구조가 복잡하기 때문에 다른 언어의 인식보다 벵골어 문자에 대한 인식이 더 어렵다. 벵골 수기 데이터 셋에 대한 인식 알고리즘을 개발하기 위해서는 주어진 수기 벵골 영상에서 그래핌 루트(Grapheme root), 모음 디아크리틱스(Vowel diacritic), 자음 디아크리틱스(Consonant diacritic) 세 가지 성분을 개별적으로 분류해야 했다. 본 연구에서는 이러한 벵골 수기 데이터 세트에 대해 세 개의 출력을 갖는 branch를 갖는 모델을 기반으로 GhostNet, EfficientNet, SENet 등 세 가지 backbone 아키텍처를 이용하여 인식을 수행하였다. 나아가 Mixup, Cutout, Cutmix, GridMask 등 4가지 데이터 증강 방법에 대해 비교 분석하여 인식률 향상을 꾀하였다. 결론적으로 우리는 GridMask 데이터 증강 기법을 사용한 EfficientNet-B5 1개, SE-ResNeXt-50 2개를 이용한 앙상블 네트워크를 기반으로 93.74%의 최고 정확도를 달성하였으며, 이 결과는 Kaggle 벵갈리 대회에 참가하는 2,059개 팀 중 상위 3.1%에 해당한다. Recently, a new dataset for a Bengali recognition was released at Kaggle and an international challenge called Bengali.AI was held. In this paper, we share the results of our participation in the competition. Since Bengali character has a more complex structure than other languages, recognition of Bengali character is more challenging than the recognition of other languages. To develop the recognition algorithm on Bengali handwritten dataset, we had to classify three components individually: Grapheme root, Vowel diacritics, and Consonant diacritics from a given handwritten Bengali image. We propose a method to improve the performance of recognition for Bengali handwritten dataset in this paper. In order to compare the quality of different models and find the optimal strategy to get better accuracies, we have trained several models based on three modern architectures (GhostNet, EfficientNet, SENet). Furthermore, we have analyzed four kinds of data augmentation methods such as Mixup, Cutout, Cutmix, and GridMask. Finally, we have achieved the best accuracy of 93.74% based on an ensemble network using one EfficientNet-B5 and two SE-ResNeXt-50 with GridMask data augmentation, and this result is the top 3.1% of the 2,059 teams participating in the Kaggle Bengali.AI challenge.