image3 [논문 리뷰] Palette: Image-to-Image Diffusion Models [논문]Palette: Image-to-Image Diffusion ModelsCitations: 1,543https://arxiv.org/abs/2111.05826IntroductionDenoising loss ft.$L_2$ - higher degree of diversity$L_1$ - conservative outputsU-Net architecture of Paletteself-attention layers 제거 → fully convolutional model ⇒ 성능 저하 Related workGAN - widely used but auxiliary objectives on structures, context, edges, contours 필요diversity 저하 요인Inpainitng기존.. 2025. 3. 21. [논문 리뷰] CCDM: Continuous Conditional Diffusion Models for Image Generation [논문]CCDM: Continuous Conditional Diffusion Models for Image GenerationCoRR 2024Citations: 1[https://arxiv.org/abs/2405.03546]0.AbstractContinuous Conditional Generative Modeling (CCGM)scalar continuous variables의 conditioned인 high-dimensional data의 distribution을 estimateCcGANs - extremely sparse or imbalanced data에 취약함Conditional diffusion models (CcDPMs) - CcGAN에서 diffusion model로 변경⇒ CCGM task.. 2025. 3. 21. [개념 설명] ResNet, UNet Computer Vision 분야에서 많이 사용되는 기본 구조 ResNet과 UNet에 대해 간단히 알아볼 예정 [논문]1. ResNet - Deep Residual Learning for Image Recognitionhttps://arxiv.org/abs/1512.03385 2. UNet - U-Net: Convolutional Networks for Biomedical Image Segmentationhttps://arxiv.org/abs/1505.045971. Residual blockResidual Mapping = $F(x) + x$$F(x)$: input x가 weight layer 통과한 outputResidual BlockSkip connection: layer의 output을 몇 개의 .. 2025. 3. 11. 이전 1 다음
