Multi-Camera Tracking with Adaptive Resource Allocation

Han, Bohyung,Joo, Seong-Wook,Davis, Larry S. Springer-Verlag 2011 International journal of computer vision Vol.91 No.1

딥 러닝의 한계와 해결책

한보형(Han, Bohyung) 한국정보과학회 2019 정보과학회지 Vol.37 No.2

Small gold nanorods-loaded hybrid albumin nanoparticles with high photothermal efficacy for tumor ablation

Seo, Bohyung,Lim, Kyungseop,Kim, Sung Soo,Oh, Kyung Taek,Lee, Eun Seong,Choi, Han-Gon,Shin, Beom Soo,Youn, Yu Seok Elsevier 2019 Colloids and surfaces Biointerfaces Vol.179 No.-

<P><B>Abstract</B></P> <P>Photothermal therapy using gold nanorods (AuNRs) has gained great attention for cancer therapy because AuNRs emit heat and induce tumor cell death responding to the near infrared light. However, the anticancer efficiency of AuNRs alone is undermined by its poor in vivo stability and potential toxicity. The prime purpose of this study was to send more AuNRs into tumors to more fully ablate them. For this, we fabricated hybrid albumin nanoparticles encapsulating small AuNRs (AuNRs-Alb-NPs), which take advantage of biocompatible albumin as a carrier, with better tumor targetability and high in vivo photothermal activity. The sizes of length/width of AuNRs were approximately 20.5 nm and 4.6 nm, respectively, showing a 4.5 aspect ratio, and the size of the resulting AuNRs-Alb-NPs was ˜130 nm, all of which are favorable for glomerular filtration and passive tumor targeting via extravasation. We chose the best formulation for AuNRs-Alb-NPs by in vitro cytotoxicity based on photothermal conversion efficiency considering the incorporated number of AuNRs. Visualized by a photothermal camera, the local tumor temperature of mice treated with AuNRs-Alb-NPs increased to 57℃, which was sufficient for the hyperthermal effect with 808 nm laser irradiation. Subsequently, AuNRs-Alb-NPs displayed remarkably better tumor ablation vs. naïve formulation of AuNRs (tumor volume: 73.8 ± 105.8 vs. 1455.3 ± 310.4 mm<SUP>3</SUP> at day 8) in the glioblastoma N2a tumor-bearing mice. Most of all, we demonstrated, using photoacoustic imaging and inductively coupled plasma mass spectrometry, that this much better tumor ablation was due to enhanced tumor targeting with albumin nanoparticles. We believe our AuNRs-Alb-NPs should be considered promising photothermal agents that are safer, have good targetability, and exhibit excellent tumor ablation.</P> <P><B>Highlights</B></P> <P> <UL> <LI> We developed a prototype of small gold nanorods-loaded albumin nanoparticles. </LI> <LI> The nanoparticles exhibited noticeable cytotoxicity on the glioblastoma N2a cells. </LI> <LI> The nanoparticles were much more potent in killing N2a tumors in vivo versus naïve gold nanorods. </LI> <LI> The hybrid albumin nanoparticles would be a potential photothermal agent for treating brain cancer. </LI> </UL> </P> <P><B>Graphical abstract</B></P> <P>[DISPLAY OMISSION]</P>

Extended Support Vector Machines for Object Detection and Localization

Jan Feyereisl,Bohyung Han 대한전자공학회 2012 전자공학회지 Vol.39 No.2

비디오에서 효율적인 행동 인식을 위한 동적 계층 생략

선종현(Jonghyeon Seon),한보형(Bohyung Han) 대한전자공학회 2022 대한전자공학회 학술대회 Vol.2022 No.11

비디오 콘텐츠(예: 인간 행동이 담긴 내용) 분석의 어려움 중 하나는 긴 비디오를 처리해야 하는 연산에서 발생하는 높은 계산 비용입니다. 우리는 프레임간의 관련성과 중요성에 따라 개별 프레임에 대해 적합하게 계산 자원을 할당하는 효율적인 동작 인식 알고리즘을 제안합니다. 특히, 본 논문의 알고리즘은 LSTM 기반 정책 모듈을 적용하고 심층 신경망 중간 계층들에서의 추출되는 특징을 기반으로 각 프레임의 유용성을 순차적으로 추정합니다. 특정 프레임이 동작을 인식하는 데 도움이 되지 않을 경우 모델은 나머지 계층으로 기능을 전달하는 것을 중단하고 다음 프레임을 처리합니다. 추론 과정 중에 간단하면서도 효과적인 조기 종료 전략을 도입하여 접근 방식의 계산 비용을 더욱 절감합니다. 제안된 알고리즘을 ActivityNet-v1.3, Mini-Kinetics 의 두 가지 공개 벤치마크에서 평가합니다. 우리의 실험은 제안된 접근 방식이 행동 인식에서 정확성과 효율성 사이의 뛰어난 균형을 달성한다는 것을 보여줍니다.

설명 문장 생성을 통한 해석 가능한 시각적 질의응답 모델 분석

김단일(Danil Kim),한보형(Bohyung Han) 한국컴퓨터정보학회 2020 한국컴퓨터정보학회 학술발표논문집 Vol.28 No.2

본 연구에서는 설명 문장 생성을 통한 해석 가능한 시각적 질의응답 모델을 설계하고 학습 방법을 제시한다. 설명 문장은 시각적 질의응답 모델이 응답을 예측하는 데에 필요한 이미지 및 질문 정보와 적절한 논리적인 정보의 조합 및 정답 추론 과정이 함의되어 있을 것으로 기대한다. 설명 문장 생성 과정이 포함된 시각적 질의응답의 기본적인 모델을 기반으로 여러 가지 학습방법을 통해 설명 문장 생성 과정과 응답 예측 과정간의 상호관계를 분석한다. 이러한 상호작용을 적극적으로 활용할 수 있는 보다 개선 시각적 질의응답 모델을 제안한다. 또한 학습한 결과를 바탕으로 설명 문장의 특성을 활용하여 시각적 질의응답 추론 과정을 개선함으로써 시각적 질의응답 모델의 발전 방향을 논의한다. 본 실험을 통해서 응답 예측에 적절한 설명 문장을 제시하는 해석 가능한 시각적 질의응답 모델을 제공한다.

수평 분할 윈도우를 이용한 컬러 객체간 폐색 탐지

조아라(Ahra Jo),한보형(Bohyung Han),장길진(Gil-Jin Jang) 한국HCI학회 2012 한국HCI학회 학술대회 Vol.2012 No.1

본 논문에서는 폐색이 존재하는 연속된 컬러 이미지에서의 객체를 추적하는 효율적인 방법을 제안한다. 이를 위하여 추적하고자 하는 물체를 포함하는 윈도우를 수평으로 분할하고, 각 윈도우의 색상 분포도의 거리의 변화에 따라 객체 사이의 폐색 패턴을 탐지하는 방법을 제안한다. 이는 대부분의 객체들이 수직 보다는 수평으로 움직인다는 가정에 따른다. 추적하고자 하는 물체를 6개의 동일한 폭을 가지는 윈도우로 분할한 뒤, 윈도우 사이의 거리변화량의 최대값과 최소값의 관계에 따라 폐색 패턴을 결정한다. 각각의 윈도우의 폐색 여부는 컬러 히스토그램 일치 확률 계산에 반영되고, 기존의 파티클 필터(particle filter) 기반의 물체 추적 알고리즘에 적용하였다. 실제 촬영된 연속 이미지에 적용된 결과 제안된 방법은 폐색 영역을 잘 탐지하였으며, 폐색이 존재하는 이미지열에 대하여 물체 추적 성능을 향상시킬 수 있었다. This paper proposes a practical algorithm for detecting occlusions in a color image sequence based on matching color histograms of horizontally segmented rectangular windows. Based on the empirical observation that the objects are more likely to move horizontally than vertically, the target window is divided into six equi-width sub-regions, and the proposed method determines the current occlusion pattern based on maximum and minimum forward difference of sub-region histogram distance. The partial occlusion detection results are embedded into likelihood computation of the conventional tracking algorithm using particle filters. Experimental results with real scenes show that the proposed method detects the occluded region quite well and improves the performance of the conventional object tracking method.

Albumin nanoparticles with synergistic antitumor efficacy against metastatic lung cancers

Kim, Bomi,Seo, Bohyung,Park, Sanghyun,Lee, Changkyu,Kim, Jong Oh,Oh, Kyung Taek,Lee, Eun Seong,Choi, Han-Gon,Youn, Yu Seok Elsevier 2017 Colloids and surfaces Biointerfaces Vol.158 No.-

<P><B>Abstract</B></P> <P>Albumin nanoparticles are well-known as effective drug carriers used to deliver hydrophobic chemotherapeutic agents. Albumin nanoparticles encapsulating curcumin and doxorubicin were fabricated using slightly modified nanoparticle albumin-bound (nab™) technology, and the synergistic effects of these two drugs were examined. Albumin nanoparticles encapsulating curcumin, doxorubicin, and both curcumin and doxorubicin were prepared using a high pressure homogenizer. The sizes of albumin nanoparticles were ∼130nm, which was considered to be suitable for the EPR (enhanced permeability and retention) effect. Albumin nanoparticles gradually released drugs over a period of 24h without burst effect. To confirm the synergistic effect of two drugs, <I>in vitro</I> cytotoxicity assay was performed using B16F10 melanoma cells. The cytotoxic effect on B16F10 melanoma cells was highest when co-treated with both curcumin and doxorubicin compared to single treatment of either curcumin and doxorubicin. The combined index calculated by medium-effect equation was 0.6069, indicating a synergistic effect. Results of confocal laser scanning microscopy and fluorescence-activated cell sorting corresponded to results from an <I>in vitro</I> cytotoxicity assay, indicating synergistic cytotoxicity induced by both drugs. A C57BL/6 mouse model induced by B16F10 lung metastasis was used to study <I>in vivo</I> therapeutic effects. When curcumin and doxorubicin were simultaneously treated, the metastatic melanoma mass in the lungs macroscopically decreased compared to curcumin or doxorubicin alone. Albumin nanoparticles encapsulating two anticancer drugs were shown to have an effective therapeutic result and would be an excellent way to treat resistant lung cancers.</P> <P><B>Highlights</B></P> <P> <UL> <LI> We developed curcumin and doxorubicin co-loaded albumin nanoparticles. </LI> <LI> The nanoparticles were fabricated by high pressure homogenization <I>via</I> Nab<SUP>®</SUP> technique. </LI> <LI> The nanoparticles exhibited synergistic cytotoxicity on the B16F10 cells. </LI> <LI> These new albumin nanoparticles would be potential as a treatment for metastatic lung cancers. </LI> </UL> </P> <P><B>Graphical abstract</B></P> <P>[DISPLAY OMISSION]</P>

Weakly Supervised Learning with Deep Convolutional Neural Networks for Semantic Segmentation: Understanding Semantic Layout of Images with Minimum Human Supervision

Hong, Seunghoon,Kwak, Suha,Han, Bohyung IEEE 2017 IEEE signal processing magazine Vol.34 No.6

<P>Semantic segmentation is a popular visual recognition task whose goal is to estimate pixel-level object class labels in images. This problem has been recently handled by deep convolutional neural networks (DCNNs), and the state-of-the-art techniques achieve impressive records on public benchmark data sets. However, learning DCNNs demand a large number of annotated training data while segmentation annotations in existing data sets are significantly limited in terms of both quantity and diversity due to the heavy annotation cost. Weakly supervised approaches tackle this issue by leveraging weak annotations such as image-level labels and bounding boxes, which are either readily available in existing large-scale data sets for image classification and object detection or easily obtained thanks to their low annotation costs. The main challenge in weakly supervised semantic segmentation then is the incomplete annotations that miss accurate object boundary information required to learn segmentation. This article provides a comprehensive overview of weakly supervised approaches for semantic segmentation. Specifically, we describe how the approaches overcome the limitations and discuss research directions worthy of investigation to improve performance.</P>

흡수 마르코프 체인을 활용한 약지도 기반 비디오 내 이상 사건 검출

김준하(Junha Kim),한보형(Bohyung Han) 한국정보과학회 2020 한국정보과학회 학술발표논문집 Vol.2020 No.12