본문 바로가기

논문 리뷰/딥러닝11

(작성 중...)[JIS] Deep Learning for Image Classification [JIS] Journal Introduction Summary: Deep Learning for Image Classification [Since 2017] 1. [EfficientNet] Tan, Mingxing, and Quoc Le. "Efficientnet: Rethinking model scaling for convolutional neural networks." International conference on machine learning. PMLR, 2019. 2. [NasNet] Zoph, Barret, et al. "Learning transferable architectures for scalable image recognition." Proceedings of the IEEE con.. 2023. 12. 26.
(작성 중...)[JIS] Deep Learning for Image Segmentation [JIS] Journal Introduction Summary: Deep Learning for Image Segmentation [Since 2020] 1. Zhou, Man, et al. "Deep fourier up-sampling." arXiv preprint arXiv:2210.05171 (2022). 2. (Swin-UNet) Cao, Hu, et al. "Swin-unet: Unet-like pure transformer for medical image segmentation." European conference on computer vision. Cham: Springer Nature Switzerland, 2022. 2. (Swin UNETR) Hatamizadeh, Ali, et al.. 2023. 11. 25.
[FCN] Long et al., 2015, Fully Convolutional Networks for Semantic Segmentation # 세줄 요약 # 저자들이 소개한 FCN은 엔드 투 엔드(end-to-end), 픽셀에서 픽셀로(pixels-to-pixels) 학습되어, 즉 입력으로도 '영상'이 들어가고, 출력에서도 분할된 '영상'이 나오는 시멘틱 분할(semantic segmentation)에서 가장 높은 성능을 보인 합성곱 신경망이다. FCN의 핵심은 네트워크 이름에도 들어가 있듯이 'Fully convolutional Network'(완전 연결된 합성곱 신경망) 구조가 핵심 아이디어이며, 이를 구현하기 위해서 기존의 분류에 사용된 합성곱신경망 모델인 AlexNet, VGG, GoogLeNet 등을 기본모델(baseline model)로 사용하고, 이들 모델의 학습된 웨이트(weight)를 미세조정(fine-tuning)하여 분할.. 2022. 11. 24.
[Review] Minaee et al., 2021, Image Segmentation Using Deep Learning: A Survey (* 작성 중; ~2022.11.04) # 세줄 요약 # 이미지 분할(Image segmentation)은 컴퓨터 비전(computer vision)과 영상처리(image processing)에서 가장 중요한 업무 중의 하나로 장면 이해, 의료 영상 분석, 로봇 인식, 비디오 감시, 증강 현실, 이미지 압축 등의 다양한 분야에서 여러 분할 알고리즘들이 선행 연구 문헌에서 발견된다. 저자들은 최근 문헌들을 뒤져서 픽셀 단위에서 작동하는 semantic and instance segmentation 합성곱 신경망, 인코더-디코더 구조, 다양한 스케일에서 피라미드 구조 접근법, 순환 신경망, 시각적 집중 모델(visual attention models), 적대적 생성 신경망(generative models i.. 2022. 10. 21.
[U-Net] Ronneberger et al., 2015, U-Net: Convolutional Networks for Biomedical Image Segmentation. # Three-line Summary # We present a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently. The architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localization. Using the network trained on transmitted light microscopy images, we won the ISBI cell trac.. 2022. 8. 11.
[Inception V3] Szegedy et al., 2016, Rethinking the Inception Architecture for Computer Vision # 세줄 요약 # Since 2014 very deep convolutional networks started to become mainstream, yielding substantial gains in various benchmarks. We are exploring ways to scale up networks in ways that aim at utilizing the added computation as efficiently as possible by suitably factorized convolutions and aggressive regularization. We benchmark our methods on the ILSVRC 2012 classification challenge valida.. 2022. 4. 11.
[MobileNet] Howard et al., 2017, MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications # 세줄 요약 # MobileNets are based on a streamlined architecture that uses depthwise separable convolutions to build light weight deep neural networks. We introduce two simple global hyper-parameters that efficiently trade off between latency and accuracy. We present extensive experiments on resource and accuracy tradeoffs and show strong performance compared to other popular models on ImageNet clas.. 2022. 3. 21.
[DenseNet] Huang et al., 2017, Densely Connected Convolutional Networks # 세줄 요약 # We introduce the Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion. For each layer, the feature-maps of all preceding layers are used as inputs, and their own feature-maps are used as inputs into all subsequent layers. DenseNet has several compelling advantages: they alleviate the vanishing-gradient problem, strengthen feat.. 2022. 3. 13.
[ResNet] He et al., 2015, Deep Residual Learning for Image Recognition # 세줄 요약 # We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth (evaluate residual nets with a depth of up to 152 layers). This result won the 1st place on the.. 2021. 10. 17.
[GoogLeNet] Szegedy et al., 2015, Going Deeper with Convolutions # 세줄요약 # 'Inception'이라고 불리는 새로운 모듈로 구성한 Deep CNN(Convolutional Neural Network), 일명 'GoogLeNet' 을 처음으로 소개한 논문으로, GoogleNet은 ILSVRC14(ImageNet Large-Scale Visual Recoginition Challenge 2014)에서 Classification과 Detection 모두에서 최고의 성능을 보였다. Inception 모듈은 multi-scale processing과 Hebbian principle에서 영감을 얻어 이전 층의 특성지도(feature map)를 다양한 크기의 필터(1x1, 3x3, 5x5, pooling)로 병렬처리한 이후 다시 하나의 출력으로 합치는 구조를 가지고 있다. .. 2020. 12. 16.

티스토리툴바