• Annual Conference of KIPS
• /
• 2019.05a
• /
• pp.482-484
• /
• 2019

Learning and analyzing 3D point clouds with deep networks is challenging due to the limited and irregularity of the data. In this paper, we present a data-driven point cloud augmentation technique. The key idea is to learn multilevel features per point and to reconstruct to a similar point set. Our network is applied to a projection loss function that encourages the predicted points to remain on the geometric shapes with a particular target. We conduct various experiments using ShapeNet part data to evaluate our method and demonstrate its possibility. Results show that our generated points have a similar shape and are located closer to the object.

• Annual Conference of KIPS
• /
• 2019.05a
• /
• pp.472-474
• /
• 2019

Because of the difference in network structure and loss function, Verification and identification models have their respective advantages and limitations for person reidentification (re-ID). In this work, we propose a multi-task network simultaneously computes the identification loss and verification loss for person reidentification. Given a pair of images as network input, the multi-task network simultaneously outputs the identities of the two images and whether the images belong to the same identity. In experiments, we analyze the major factors affect the accuracy of person reidentification. To address the occlusion problem and improve the generalization ability of reID models, we use the Random Erasing Augmentation (REA) method to preprocess the images. The method can be easily applied to different pre-trained networks, such as ResNet and VGG. The experimental results on the Market1501 datasets show significant and consistent improvements over the state-of-the-art methods.

• Annual Conference of KIPS
• /
• 2019.05a
• /
• pp.542-544
• /
• 2019

Changes in a person's health affect one's lifestyle and work activities. According to the World Health Organization (WHO), abnormal activity is growing faster in people aged 60 or more than any other age group in almost every country. This trend steadily continues and expected to increase further in the near future. Abnormal activity put these people at high risk of expected incidents since most of these people live alone. Human abnormal activity analysis is a challenging, useful and interesting problem among the researchers and its particularly crucial task in life and health care areas. In this paper, we discuss the problem of abnormal activities of old people lives alone at home. We propose Convolutional Neural Network (CNN) based model to detect the abnormal behaviors of elderlies by utilizing six simulated action data from daily life actions.

• Annual Conference of KIPS
• /
• 2019.05a
• /
• pp.522-524
• /
• 2019

In spite of a rapid development in the quality of built-in mobile cameras, their some physical restrictions hinder them to achieve the satisfactory results of digital single lens reflex (DSLR) cameras. In this work we propose an end-to-end deep learning method to translate ordinary images by mobile cameras into DSLR-quality photos. The method is based on the framework of generative adversarial networks (GANs) with several improvements. First, we combined the U-Net with DenseNet and connected dense block (DB) in terms of U-Net. The Dense U-Net acts as the generator in our GAN model. Then, we improved the perceptual loss by using the VGG features and pixel-wise content, which could provide stronger supervision for contrast enhancement and texture recovery.

• Annual Conference of KIPS
• /
• 2019.05a
• /
• pp.519-521
• /
• 2019

자율주행 자동차에서의 보행자 인식 및 사람의 행동 인식과 같은 분야 등에 대한 연구들이 활발하게 진행되고 그에 기반을 둔 기술들이 많이 개발되고 있다. 그리고 대부분의 연구에서는 사람에 대한 경계 박스를 검출한다. 영상에서 사람의 유무 혹은 위치를 판단하는 문제에서는 경계 박스만을 검출하는 것이 효율적일 수 있으나 경계 박스는 행동 인식과 같은 분야에 사용하기에는 많은 정보의 손실이 발생할 수 있다. 본 논문에서는 U-NET 구조의 딥러닝 모델을 사용해 경계 박스로 인한 정보 손실을 줄일 수 있는 보행자 분할 방법을 제안한다. 모델의 학습을 위해 2017 COCO 데이터셋의 사람 카테고리를 사용하였으며 Penn-Fudan 보행자 데이터셋을 이용하여 제안 방법을 테스트하였으며 기존의 방법들과 비교하여 의미 있는 결과를 얻었다.

• Annual Conference of KIPS
• /
• 2012.04a
• /
• pp.146-148
• /
• 2012

최근 GPGPU를 활용한 병렬처리가 각광을 받고 있는 가운데 GPU의 구조적 특성인 매니코어(many core)기반에서 쓰레드(thread)의 구성이 성능에 얼마나 영향을 미치는지에 관해 수치적 해답을 얻고자 하였다. 이는 멀티코어 (multi core)기반으로 작성된 프로그램을 GPGPU로 변환하는 과정에서 쓰레드의 최대활용도를 빠르게 추측 할 수 있도록 도움을 얻고자 하는데 일차적인 목적이 있다. 현재 GPGPU의 쓰레드 구성은 입력되는 데이터의 양을 고려하여 충분한 테스트를 거쳐 경험적인 최적화 수를 지정해 주워야 한다. 이번 연구를 통해 GPGPU로 변환하는 과정에서 최적의 쓰레드 수구성 방법을 추측 할 수 있으며 더 나아가 동적으로 최적의 수를 구할 수 있도록 하는데 목적이 있다.

• Annual Conference of KIPS
• /
• 2010.04a
• /
• pp.405-408
• /
• 2010

This paper presents a new feature extraction technique, correlation coefficient and Manhattan distance (MD) based method for recognition of rotated object in an image. This paper also represented a new concept of intensity invariant. We extracted global features of an image and converts a large size image into a one-dimensional vector called circular feature vector's (CFVs). An especial advantage of the proposed technique is that the extracted features are same even if original image is rotated with rotation angles 1 to 360 or rotated. The proposed technique is based on fuzzy sets and finally we have recognized the object by using histogram matching, correlation coefficient and manhattan distance of the objects. The proposed approach is very easy in implementation and it has implemented in Matlab7 on Windows XP. The experimental results have demonstrated that the proposed approach performs successfully on a variety of small as well as large scale rotated images.

• Annual Conference of KIPS
• /
• 2021.11a
• /
• pp.647-650
• /
• 2021

The most powerful and modern face recognition techniques are using deep learning methods that have provided impressive performance. The outbreak of COVID-19 pneumonia has spread worldwide, and people have begun to wear a face mask to prevent the spread of the virus, which has led existing face recognition methods to fail to identify people. Mainly, it pushes masked face recognition has become one of the most challenging problems in the face recognition domain. However, deep learning methods require numerous data samples, and it is challenging to find benchmarks of masked face datasets available to the public. In this work, we develop a new simulated masked face dataset that we can use for masked face recognition tasks. To evaluate the usability of the proposed dataset, we also retrained the dataset with ArcFace based system, which is one the most popular state-of-the-art face recognition methods.

• Annual Conference of KIPS
• /
• 2021.11a
• /
• pp.856-859
• /
• 2021

Image transmission by means of telecommunications is an essential task for information sharing. For considerable distances, wireless channels can be utilized and tuned for proper uses of image data exchange. However, the disturbances that a radio wave encounter during transmission causes partial or total loss of information. Result of such communications is a distorted image at the receiver's end. This paper proposes an auto-encoder architecture as an image enhancement method for narrow-bandwidth radio images. With this method, a distorted image can be improved for better receiver satisfaction. The proposed auto-encoder is trained with many narrow-bandwidth radio image data; hence it enhances a given distorted image. Also, the results were verified with the original image data being the reference images.

• Annual Conference of KIPS
• /
• 2021.11a
• /
• pp.878-879
• /
• 2021

Most state-of-the-art CNNs for action recognition are based on a two-stream architecture: RGB frames stream represents the appearance and the optical flow stream interprets the motion of action. However, the cost of optical flow computation is very high and then it increases action recognition latency. We introduce a design strategy for action recognition inspired by a two-stream network and teacher-student architecture. There are two sub-networks in our neural networks, the optical flow sub-network as a teacher and the RGB frames sub-network as a student. In the training stage, we distill the feature from the teacher as a baseline to train student sub-network. In the test stage, we only use the student so that the latency reduces without computing optical flow. Our experiments show that its advantages over two-stream architecture in both speed and performance.