Computer Vision116 [2025-1] 정인아 - Image Super-Resolution via Iterative Refinement https://arxiv.org/abs/2104.07636 Image Super-Resolution via Iterative RefinementWe present SR3, an approach to image Super-Resolution via Repeated Refinement. SR3 adapts denoising diffusion probabilistic models to conditional image generation and performs super-resolution through a stochastic denoising process. Inference starts with parxiv.org Intro문제기존 GAN 기반 super-resolution 모델은 보기에 그럴듯해보이고, 실.. 2025. 2. 1. [2025-1] 유경석 - Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference (LCM) https://arxiv.org/pdf/2310.04378 https://blog.outta.ai/171 [2025-1] Latent Consistency Models : Synthesizing High-Resolution ImagesWith Few-Step Inference논문 링크: 2310.04378 참고 논문 리뷰 블로그 링크: Latent Consistency Models : Synthesizing High-Resolution ImagesWith Few-Step Inference 논문 리뷰 :: LOEWEN Latent Consistency Models : Synthesizing High-Resolution ImagesWith Few-Stblog.outta.ai SummaryStable Dif.. 2025. 2. 1. [2025-1] 주서영 - SRDiff : Single image super-resolution with diffusion probabilistic models SRDiff SRDiff: Single Image Super-Resolution with Diffusion Probabilistic ModelsSingle image super-resolution (SISR) aims to reconstruct high-resolution (HR) images from the given low-resolution (LR) ones, which is an ill-posed problem because one LR image corresponds to multiple HR images. Recently, learning-based SISR methods have garxiv.orgNeurocomputing 2021611회 인용※ 참고SR3 Image Super-Resolut.. 2025. 2. 1. [2025-1] 전윤경- Roformer: Enhanced Transformer with Rotary Position Embedding IntroductionRoPE: 회전 행렬을 사용하여 절대적인 위치를 인코딩하고 self attention 공식에 명시적인 상대적 위치 의존성을 통합함.유연한 시퀀스 길이 지원상대적 거리가 증가함에 따라 토큰 간의 의존도 감소linear self-attention 메커니즘에서도 상대적 위치 인코딩을 적용할 수 있는 능력 갖춤Roformer: 회전 위치 임베딩(RoPE)을 적용한 Transformer 모델 -> 기존 방법보다 우수한 성능 Background and Related Workpreliminary$ S_{N}=\left\{w_{i} \right\}_{i=1}^{N}$ : N개 인풋 토큰의 시퀀스.$ E_{N}=\left\{x_{i} \right\}_{i=1}^{N}$ : PE 가 적용되지.. 2025. 1. 31. 이전 1 ··· 8 9 10 11 12 13 14 ··· 29 다음
