목록Review (Paper, Code ...)/Paper Review
반응형
(6)
s5unnyjjj's LOG
Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising
paper : ieeexplore.ieee.org/abstract/document/7839189 1. Introduction 2. The Proposed Denoising CNN model 3. Experimental Results 4. Conclusion ---------------------------------------------------------------------------------------------------- 1. Introduction - Cscaed of Shrinkage Fields (CSF)와 Trainable Nonlienar Reaction Diffusion (TNRD) 같은 이전의 디노이즈 모델은 certain noise level만을 위한 학습..
Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
paper: openaccess.thecvf.com/content_iccv_2017/html/Zhu_Unpaired_Image-To-Image_Translation_ICCV_2017_paper.html 1. Introduction 2. Related Work 3. Formulation 4. Network Architecture 5. Results 6. Limitations 7. Conclusion ---------------------------------------------------------------------------------------------------- 1. Introduction - CycleGAN의 경우에는 전체적인 형태를 유지하는 Style Transfer..
Image-to-Image Translation with Conditional Adversarial Networks
paper: openaccess.thecvf.com/content_cvpr_2017/html/Isola_Image-To-Image_Translation_With_CVPR_2017_paper.html 1. Introduction 2. Formulation 3. Network Architecture 4. Result 5. Conclusion ---------------------------------------------------------------------------------------------------- 1. Introduction - Image to Image Mapping Network에서는 Photo-realistic을 추구한다. CNN은 모든 출력의 평균을 최소화하..
DeblurGAN: Blind Motion Deblurring Using Conditional Adversarial Network
paper : openaccess.thecvf.com/content_cvpr_2018/html/Kupyn_DeblurGAN_Blind_Motion_CVPR_2018_paper.html 1. Introduction 2. The proposed method 3. Result 4. Conclusion ---------------------------------------------------------------------------------------------------- 1. Introduction DeblurGAN은 Deblurred Image를 생성하여 object detection에 도움을 준다. Discriminator로 Wasserstein GAN을 사용하였으며 그외에도 ..
Densely Connected Convolutional Networks[DenseNet]
paper : arxiv.org/abs/1608.06993 1. Introduction 2. DenseNet 3. DenseNet-BC 4. Result 5. Concolusion ---------------------------------------------------------------------------------------------------- 1. Introduction 네트워크의 깊이가 깊어질수록 초기의 정보가 사라진다는 문제점이 있다. 그래서 일각에서는 해당 문제를 해결하기위해 노력하고 있다. 예를 들면 Stochastic depth는 레이어를 무작위로 dropping하여 ResNet을 단축시킴으로써 초기 정보를 밑으로 더 많이 내려가게 하고 FractalNets..
Conditional Generative Adversarial Nets[CGAN]
paper: arxiv.org/abs/1411.1784 1. Introduction 2. Generative Adversarial Nets 3. Conditional Adversarial Nets 4. Result 5. Conclusion ---------------------------------------------------------------------------------------------------- 1. Introduction Unconditioned generative model은 생성되는 데이터에 어떠한 제어가 없다. 반면 Conditional Adversarial Net은 Generative Discriminator에 condition 적용함으로써 Additi..