최근 탑티어 학회에 diffusion model 관련된 논문의 양이 매우 증가했습니다.
CNN에서 GAN으로, Transformer가 나오나 싶더니 Diffusion Model이 SOTA를 위협하고 있습니다.
Diffusion model에 입문부터 논문까지 도달하기 위한 path를 정리해보았습니다.
Diffusion Model 논문의 흐름
Diffusion Model의 큰 축이되는 대표적인 논문의 흐름은 다음과 같습니다.
•
[ DPM, Diffusion Probabilistic models ] Jascha Sohl-Dickstein et al.
“Deep Unsupervised Learning using Nonequilibrium Thermodynamics.”
ICML 2015.
•
[ NCSN, Noise-conditioned score network ] Yang Song & Stefano Ermon.
“Generative modeling by estimating gradients of the data distribution.”
NeurIPS 2019.
•
[ NCSNv2 ] Yang Song & Stefano Ermon.
“Improved techniques for training score-based generative models.”
NeuriPS 2020.
•
[ DDPM ] Jonathan Ho et al.
“Denoising diffusion probabilistic models.”
arxiv Preprint arxiv (2020). [code]
•
[ DDIM ] Jiaming Song et al.
“Denoising diffusion implicit models.”
arxiv Preprint arxiv (2020). [code]
•
[ IDDPM ] Alex Nichol & Prafulla Dhariwal.
“Improved denoising diffusion probabilistic models”
arxiv Preprint arxiv (2021). [code]
•
[ ADM ] Prafula Dhariwal & Alex Nichol.
“Diffusion Models Beat GANs on Image Synthesis."
arxiv Preprint arxiv (2021). [code]
•
[ NCSN++ ] Yang Song et al.,
“Score-Based Generative Modeling through Stochastic Differential Equations”
ICLR 2021 (Outstanding Paper Award)
•
[ ScoreFlow ] Yang Song*, Conor Durkan*, Iain Murray, and Stefano Ermon.
”Maximum Likelihood Training of Score-Based Diffusion Models”
NeurIPS 2021 (Spotlight)
•
[ LDM a.k.a stableDiffusion ] Rombach, Robin, et al.
"High-resolution image synthesis with latent diffusion models."
CVPR 2022, Oral [code]
What are Diffusion Models?
Diffusion Model을 깔끔하게 정리한 블로그입니다.
Score-Based Diffusion Models
Generative Modeling by Estimating Gradients of the Data Distribution
This blog post focuses on a promising new direction for generative modeling. We can learn score functions (gradients of log probability density functions) on a large number of noise-perturbed data distributions, then generate samples with Langevin-type sampling.
https://yang-song.net/blog/2021/score/
Generative Modeling by Estimating Gradients of the Data Distribution by Yang Song, May 5, 2021
시간이 흘러 Score-based Diffusion model 이 나옵니다.
많은 애니메이션과 코드가 포함되어 있는 포스트입니다.
Score-Based Diffusion Models를 한국어로 설명한 고려대학교 김성범 소장님의 강의입니다.
UNIST 인공지능 대학원 유재준 교수님의 한국어 논문 세미나 영상입니다.
lil log 블로그 글을 읽으신 후 들어보시는걸 추천드립니다.
[ADM] Diffusion Model Beats GAN
Classifier-Free Diffusion Guidance
[PITI] Pretraining is All You Need for Image-to-Image Translation
•
사전 훈련된 확산 모델을 통한 image-to-image 변환
Cascaded Diffusion Models for High Fidelity Image Generation
•
Cascade 구조를 통해 확산 모델에서 고해상도 샘플 생성