일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
1 | 2 | 3 | 4 | |||
5 | 6 | 7 | 8 | 9 | 10 | 11 |
12 | 13 | 14 | 15 | 16 | 17 | 18 |
19 | 20 | 21 | 22 | 23 | 24 | 25 |
26 | 27 | 28 | 29 | 30 | 31 |
- 사회초년생 추천독서
- paper review
- 논문 분석
- InstructPix2Pix
- DACON
- Git
- TensorFlow Object Detection API install
- CARLA simulator
- TensorFlow Object Detection Error
- 크롤링
- Branch 활용 개발
- DOTA dataset
- Custom Animation
- object detection
- AI Security
- Paper Analysis
- Linux build
- Towards Deep Learning Models Resistant to Adversarial Attacks
- 논문분석
- TensorFlow Object Detection 사용예시
- TensorFlow Object Detection Model Build
- 객체 탐지
- Carla
- 커스텀 애니메이션 적용
- VOC 변환
- Docker
- 리눅스 빌드
- 기능과 역할
- 개발흐름
- Object Detection Dataset 생성
- Today
- Total
목록분류 전체보기 (47)
JSP's Deep learning
1. 요약 YOLOv3는 Residual 기법이 적용된 DarkNet-53 구조를 사용하였다. 클래스 분류에 Logistic classifiers를 사용하였다. K-means 클러스터링을 통해서 Bounding box priors을 구성했다. (Bounding box priors에 대해서는 뒤에서 설명한다) 3-scales feature map을 사용하여 feature을 추출한다. YOLOv3는 YOLOv2와 비교했을 때, 작은 객체에 대한 탐지 성능이 향상되었다. 2. 용어정리 1) Linear activation f(x) = cx의 식을 가지는 선형 활성화 함수 다중 출력이 가능하다. 미분 값이 상수이기 때문에 오차역전파를 통한 학습이 불가능하다는 특징이 존재한다. 2) Upsampling Decon..
박지성(Park Ji Seung) wjsrlahrlco@daum.net Pukyong National University - Computer Engineering GitHub : https://github.com/wjsrlahrlco1998 2022.03.14 ~ 2022.08.11 삼성 멀티캠퍼스 - 서비스 산업 데이터를 활용한 빅데이터 분석 과정 수료 - 개인 최우수상 2022.05.30 ~ 2022.06.10 DACON - 중고차 가격 예측 경진대회(53 / 390) (https://dacon.io/competitions/official/235901/overview/description) 2022.07.01 ~ 2022.08.05 DACON - 신용카드 사기 거래 탐지 AI 경진대회(7 / 740)..
1. 요약 ①“Optimization 관점”에서 Adversarial Robustness 정의 (=> 강력한 Adversarial Attack에 대한 Defense을 학습) ②Adversarial Training with Strong Attack => Robustness↑ ③ Model’s Capacity↑=> 복잡하고 정교한 Decision boundary => Robustness↑ ④복잡한 Dataset에 대한 Attack에 대해서는 Defense 성능이 좋지 못하다. 2. 용어 정리 1)“First-Order Attack” “한 번의 미분”만으로 공격을 수행하는 기법(Loss function에 대해서 한 번) => 즉, 오차역전파를 한번만 수행함. PGD(Projected Gradient Desce..
CARLA UE4Editor crash 오류 수정 중 : 23.02.20 ~ 23.02.21(수정완료) * 메인 PC 사전준비 사항 NVIDIA-Docker 및 Docker Container Toolkit 설치 NVIDIA 그래픽 드라이버 및 CUDA, cuDNN 설치 (즉, GPU 사용가능 환경 구성) Vulkan SDK 환경 구성 (참고 : https://vulkan.lunarg.com/doc/view/latest/linux/getting_started.html) nvidia/vulkan 이미지 빌드 (참고 : https://jsp-coding.tistory.com/43) xhost 권한 설정 # 다음의 명령어로 권한 설정(모든 권한 허용) xhost + * 빌드환경 메인 PC OS : Ubuntu ..
1. Dockerfile 다운로드 git clone https://gitlab.com/nvidia/container-images/vulkan.git 2. 이미지 빌드 docker build --pull -t \ nvidia/vulkan:1.3-470 \ --build-arg BASE_DIST=ubuntu20.04 \ --build-arg CUDA_VERSION=11.4.2 \ --build-arg VULKAN_SDK_VERSION=`curl -sk https://vulkan.lunarg.com/sdk/latest/linux.txt` \ --file docker/Dockerfile.ubuntu . 3. 컨테이너 생성 docker run -itd --gpus all --name test -e NVIDIA_D..
CARLA UE4Editor crash 오류 수정 중 : 23.02.20~23.02.21(수정완료) * 메인 PC 사전준비 사항 NVIDIA-Docker 및 Docker Container Toolkit 설치 NVIDIA 그래픽 드라이버 및 CUDA, cuDNN 설치 (즉, GPU 사용가능 환경 구성) Vulkan SDK 환경 구성 (참고 : https://vulkan.lunarg.com/doc/view/latest/linux/getting_started.html) nvidia/vulkan 이미지 빌드 (참고 : https://jsp-coding.tistory.com/43) xhost 권한 설정 # 다음의 명령어로 권한 설정(모든 권한 허용) xhost + * 빌드환경 메인 PC OS : Ubuntu 20..
1. 요약 YOLO9000은 9000개 이상의 categories에 대해서 detection이 가능한 object detection system이다. (YOLOv1은 200-classes) YOLOv2는 Speed와 accuracy의 tradeoff가 잘 절충된 model이다. YOLO9000에서는 Classification dataset과 detection dataset을 결합하여 학습하는 방법을 사용했다. 결론적으로, COCO dataset에 없는 class에 대해서도 Object detection이 가능하게 되었다. (단, 비슷한 형태의 세부 카테고리의 객체에 대해서 탐지가 가능해진 것!) 2. 주요 용어(간단 정리) ...상세한 사항은 관련 논문을 찾아봐야합니다... 1) ResNet Residu..
[CARLA] 클라이언트 개발자 환경세팅 Docker container image와 그 빌드 과정은 앞전의 글을 참고한다. * CARLA 0.9.11 : https://jsp-coding.tistory.com/38 [CARLA Simulator] CARLA 0.9.11 Linux Build with docker CARLA Simulator Linux Build on Ubuntu 20.04 using docker 본 환경구축 방법은 개발환경에 의존적이며 개발환경이 달라지면 예기치 못한 에러가 발생할 수 있다. * 개발환경 Local OS : Ubuntu 20.04 GPU : RTX 4080 NVIDIA jsp-coding.tistory.com * CARLA 0.9.13 : https://jsp-coding..