http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Kyungmin Lee,Hunmin Yang,Se-Yoon Oh 제어로봇시스템학회 2020 제어로봇시스템학회 국제학술대회 논문집 Vol.2020 No.10
Deep neural networks tend to be erroneous when the training and test distribution differ. Especially, neural classifiers are brittle to adversarial examples, and highly overconfident to out-of-distribution examples. Hybrid modeling of generative and discriminative distribution shown to be effective for out-of-distribution detection, but is not robust to adversarial attacks. Otherwise, defense methods for adversarial attacks cannot distinguish out-of-distribution examples. In this work, we present a hybrid model that can deal with both adversarial and out-of-distribution examples. Our method is built upon the joint energy based model and adversarial training. Through experiments on CIFAR-10 dataset, we show that our method has state-of-the-art performanced among hybrid models. Furthermore, we show that our model exhibits more perceptually-aligned feature than other methods, by showing the gradient sensitivity map with newly proposed score function.
Training Deep Neural Networks with Synthetic Data for Off-Road Vehicle Detection
Eunchong Kim,Kanghyun Park,Hunmin Yang,Se-Yoon Oh 제어로봇시스템학회 2020 제어로봇시스템학회 국제학술대회 논문집 Vol.2020 No.10
In tandem with growing deep learning technology, vehicle detection using convolutional neural network is now become a mainstream in the field of autonomous driving and ADAS. Taking advantage of this, lots of real image datasets have been produced in spite of the painstaking work of data collection and ground truth annotation. As an alternative, virtually generated images are introduced. This makes data collection and annotation much easier, but a different kind of problem called ‘domain gap’ is announced. For instance, in off-road vehicle detection, there is a difficulty in producing off-road image dataset not only by collecting real images, but also by synthesizing images sidestepping the domain gap. In this paper, focusing on the off-road army tank detection, we introduce a synthetic image generator using domain randomization on off-road scene context. We train a deep learning model on synthetic dataset using low level features form feature extractor pre-trained on real common object dataset. With proposed method, we improve the model accuracy to 0.86 AP@0.5IOU, outperforming naïve domain randomization approach.
Improving Instance Segmentation using Synthetic Data with Artificial Distractors
Kanghyun Park,Hyeongkeun Lee,Hunmin Yang,Se-Yoon Oh 제어로봇시스템학회 2020 제어로봇시스템학회 국제학술대회 논문집 Vol.2020 No.10
Despite the advances in deep learning, training instance segmentation models like convolutional neural networks still tend to depend on enormous training data that are expensive and require labor to annotation. To avoid labor-intensive procedure, synthetic data can be an alternative because it is easy to generate and automatically segmented. However, it is challenging to train instance segmentation model that perform well at real world using only synthetic data because of domain gap. It is wrong direction to put a lot of effort into solving these problems by making synthetic data more photorealistic. In this paper, we suggest how to learn the instance segmentation model using synthetic data with artificial distractors. The performance has been improved about 7% by adding flying distractors compared to original synthetic data.
도어트림 측면 충돌 해석 및 시험을 통한 실차 SHARP EDGE 예측에 대한 연구
김종수(Jongsoo Kim),백승현,Hunmin Park,이현철(Hunchol Lee),양희승(Heeseung Yang),김유민(Yumin Kim),강두안(Dooan Kang) 한국자동차공학회 2011 한국자동차공학회 부문종합 학술대회 Vol.2011 No.5
The sharp edge in a full-scale Car Side Impact caused injury to the drivers. It is important to prevent sharp edge of full-scale Car through a pre-verification. The pre-verification system is verified by side impact test of full-scale car. However, this test spends a lot of money and time for sharp edge of using the full-scale car. So, there are limited number of examinations. In order to solve like this problem, the method that predict sharp edge of door trim is a necessary for auto part manufactures. This paper is purposed to describes a theory of the advanced side impact test and cae to improve sharp edge of door trim. It puts emphasis on analyzing sharp edge from simplified impact test and cae to improve sharp edge based on preceding verification. The research may lead to prediction of sharp edge before full scale side impact test
Applying FastPhotoStyle to Synthetic Data for Military Vehicle Detection
Hyeongkeun Lee,Kyungmin Lee,Hunmin Yang,Se-Yoon Oh 제어로봇시스템학회 2020 제어로봇시스템학회 국제학술대회 논문집 Vol.2020 No.10
Object detection is one of the main task for the deep learning applications. Deep learning performance has already exceeded human’s detection ability, in the case when there are lots of data for training deep neural networks. In the case of military fields, there are needs to resolve the data shortage problem to employ deep learning system efficiently with benefits. Generating the synthetic data can be a solution, but the domain gap between the synthetic and real data is still an obstacle for training the model. In this paper, we propose a method for decreasing the domain gap by applying style transfer techniques to synthetic data for military vehicle detection. Utilizing FastPhotoStyle to the synthetic data aids efficiently improving the accuracy of object detection when the real data is insufficiency for training. Specifically, we show that stylization which enables artificial data more realistic diminishes the domain gap by evaluating the visualization of their distributions using principal component analysis and Fréchet inception distance score. As a result, the performance has been improved about 8% in the AP@50 metric for stylized synthetic data.