diffusion5 [논문 리뷰] Blaze3DM: Marry Triplane Representation with Diffusion for 3D Medical Inverse Problem Solving [논문]Blaze3DM: Marry Triplane Representation with Diffusion for 3D Medical Inverse Problem SolvingPreprint 2024Citations: 2https://arxiv.org/abs/2405.15241Summary[3D medical imaging reconstruction]high-dim data 특성상 computation cost 매우 큼기존 volume-wise 방식 - 전체 분포를 잘 학습하지 못해서 비일관성(artifacts) 발생일부 연구 - latent space에서 생성을 시도하지만 세부 image detail 복원이 부족[Blaze3DM]Compact triplane neural field와 powerful di.. 2025. 4. 10. [논문 리뷰] PSLD: Solving Linear Inverse Problems Provably via Posterior Sampling with Latent Diffusion Models [논문]Solving Linear Inverse Problems Provably via Posterior Sampling with Latent Diffusion ModelsNeurIPS 2023Citations:73[https://arxiv.org/abs/2307.00619]Summary기존의 DPS (Diffusion Posterior Sampling) - Pixel space에서 동작LDM (Latent Diffusion Models)로 확장하면 잘 작동 X Latent 공간에서 inverse problems을 이론적으로 정확한 posterior sampling 보장 입증PSLD - Posterior Sampling with Latent DiffusionPre-trained LDMs 사용 가능 (St.. 2025. 4. 3. [논문 리뷰] NCSN: Generative modeling by estimating gradients of the data distribution [논문]Generative Modeling by Estimating Gradients of the Data DistributionNeurIPS 2019Citations: 4,139https://arxiv.org/abs/1907.05600[references]https://yang-song.net/blog/2021/score/해당 blog를 먼저 공부하고 NCSN, SDE 논문을 보는 것 추천!!![개념 설명] Score-based Model [개념 설명] Score-based Model[Blog][https://yang-song.net/blog/2021/score/]blog 작성자 - SDE의 저명한 저자인 Yang SongNCSN, SDEs 등 Score-based model을 공부하기 전 이 blog.. 2025. 3. 25. [논문 리뷰] CcDPM: A Continuous Conditional Diffusion Probabilistic Model for Inverse Design [논문]CcDPM: A Continuous Conditional Diffusion Probabilistic Model for Inverse DesignAAAI 2024Citations: 4[https://ojs.aaai.org/index.php/AAAI/article/view/29647]0.AbstractcGANs - single-point design problems에서 promising resultsSingle-point: one performance requirement under one working conditionCategorical space 가정CcGANs - Vicinal Risk Minimization(VRM)도입했지만, 여전히 multi-point design problem 다루지 못.. 2025. 3. 21. [논문 리뷰] CG: Diffusion models beat gans on image synthesis [논문]Diffusion models beat gans on image synthesisNIPS 2021Citations: 8,013https://arxiv.org/abs/2105.052330. AbstractCG - simple, compute-efficient method for trading off diversity for fidelity using gradients from a classifier25 forward step으로 BigGAN-deep과 유사한 성능을 가지고 coverage of the distribution을 더욱 유지함 1. IntroductionGAN - image generation task의 FID, IS, Precision과 같은 sample quality metrics에서.. 2025. 3. 21. 이전 1 다음
