Skip to content

Latest commit

 

History

History
302 lines (232 loc) · 16.5 KB

diffusion.md

File metadata and controls

302 lines (232 loc) · 16.5 KB

Awesome-Diffusion-Models

This repository contains a collection of resources and papers on Diffusion Models.

If there are any missing valuable resources or papers or any materials related to diffusion model, please do not hesitate to create or pull request to issues. I am happy to reflect them.

Introductory Material

What are Diffusion Models?
Lilian Weng
2021. [Website]
11 Jul 2021

Papers

Image Generation

많은 분들이 기다리시던 저희 #NAVER #AI_Lab 의 새로운 diffusion 기반의 controllable Text-to-Image 모델 DenseDiffusion 논문과 소스코드가 공개되었습니다. 이 연구는 AI Lab 생성모델팀 김윤지님이 주저자로, 이지영님, 김진화 님, 저 그리고 CycleGAN으로 유명한 생성모델분야 글로벌 최고 연구자 중 1명인 CMU의 Jun-Yan Zhu 교수가 함께 코웍한 연구로 오는 10월 파리에서 열리는 #ICCV23 에서 발표합니다.

추가적인 훈련에 대한 오버헤드 없이 attention moderation 만으로 컨트롤이 강화되고 구체적으로 묘사된 의미의 텍스트 입력도 더욱 정확하게 이미지를 편집가능하므로 많은 분들이 활용해보시면 좋을 것 같네요. 덧으로 Jun-Yan은 학회 제출때는 물론이고 Cam-ready version의 논문 퀄리티 향상을 위해 끝까지 정말 꼼꼼하게 코멘트하고 수정하는 모습을 보여주어 왜 최고의 연구자인지 다시한번 깨닫게 해주었네요 ㅎㅎ 논문: https://arxiv.org/abs/2308.12964 github: https://github.com/naver-ai/DenseDiffusion

2022년 8월, Stable Diffusion 발표로 세상을 놀라게 했던 그때로부터 딱 1년 지났네요.

22년8월10일 Stable Diffusion 런치 발표, 22년8월22일 Stable Diffusion 모델 오픈 배포를 시작했습니다. 이전에 Dall-E 가 있었고 Midjourney 가 막 오픈베타를 시작하긴 했지만 스테이블 디퓨전의 테스트 이미지들이 꽤 놀라움을 주었습니다. 저도 Dall-E 테스트해 보고 페북에 테스트 이미지 올리기도 했었지만 막대한 비용이 들어가는 이런 LLM 이나 생성모델들이 구글등에서도 오픈하지 않는 상황에서 오픈 모델이 나올지는 생각도 못했읍니다. 이로부터 채 일주일이 지나지 않아 오픈의 힘으로 많은 곳에서 이를 활용한 것들이 쏟아지기 시작했고 SNS 와 뉴스로 들끓었습니다. "Stable Diffusion은 지금까지 나온 것중 가장 중요한 AI Art 모델" "Stable Diffusion 공개 1주일만에 벌어진 놀라운 일들" "Stability AI, 1억 1천만 달러 투자 유치" 22년 9월 인공지능 그림 미술전 1위 논란이 터지면서 더욱더 이미지 생성은 화제가 되었습니다. (이 미술전에 사용된 것은 미드저니였습니다) 오픈된 stable diffusion 을 애니메 이미지들로 파인튜닝한 waifu diffusion 등 파인튜닝 모델들이 나오기 시작하면서 더욱 다양화되고 질 좋은 수정된 수 많은 모델들이 사람들로부터 만들어져 나오기 시작했습니다. 이후 6개월 동안 기술적으로도 놀랍도록 발전되어 나갔습니다. DreamBooth, LoRA, xformers, ControlNet ... 그러나 2023년 1월에는 무단 이미지 사용 혐의로 소송이 제기되기도 하였습니다. 또한, 상용 미드저니의 놀라운 버전업과 화질 향상이 이어지는 반면 stable diffusion 2.0, 2.1 버전이 나왔지만 1.5 보다 크게 향상을 보여주지는 못했고, 사람들 활용도 덜했으며, 기본 모델들보다 오히려 사람들이 파인튜닝한 모델들이 더 많이 활용되었습니다. 2023년 7월에 고화질용 버전 SDXL 발표로 그동안 상용 미드저니 등에 품질이 떨어진다는 평을 듣던 것에서 다시 도약했다는 말을 듣기 시작했습니다. 1년만에 놀랍도록 화질이나 기술적 측면에서 발전해 왔습니다. 2023년에는 Gen-2, Pika Labs, Zeroscope 등 Text-to-Video 과 Text-to-3D 등이 나오면서 아마도 이런 것들을 준비하고 있는 것으로 보이며 추후에 이런 것들이 발표되지 않을까 싶습니다.

Bilateral Denoising Diffusion Models
Max W. Y. Lam, Jun Wang, Rongjie Huang, Dan Su, Dong Yu
arXiv 2021. [Paper] [Project]
26 Aug 2021

ILVR: Conditioning Method for Denoising Diffusion Probabilistic Models
Jooyoung Choi, Sungwon Kim, Yonghyun Jeong, Youngjune Gwon, Sungroh Yoon
ICCV 2021 (Oral). [Paper] [Github]
6 Aug 2021

SDEdit: Image Synthesis and Editing with Stochastic Differential Equations
Chenlin Meng, Yang Song, Jiaming Song, Jiajun Wu, Jun-Yan Zhu, Stefano Ermon
arXiv 2021. [Paper] [Project] [Github]
2 Aug 2021

Structured Denoising Diffusion Models in Discrete State-Spaces
Jacob Austin1, Daniel D. Johnson1, Jonathan Ho, Daniel Tarlow, Rianne van den Berg
arXiv 2021. [Paper]
7 Jul 2021

Variational Diffusion Models
Diederik P. Kingma1, Tim Salimans1, Ben Poole, Jonathan Ho
arXiv 2021. [Paper] [Github]
1 Jul 2021

Non Gaussian Denoising Diffusion Models
Eliya Nachmani1, Robin San Roman1, Lior Wolf
arXiv 2021. [Paper] [Project]
14 Jun 2021

D2C: Diffusion-Denoising Models for Few-shot Conditional Generation
Abhishek Sinha1, Jiaming Song1, Chenlin Meng, Stefano Ermon
arXiv 2021. [Paper] [Project] [Github]
12 Jun 2021

Learning to Efficiently Sample from Diffusion Probabilistic Models
Daniel Watson, Jonathan Ho, Mohammad Norouzi, William Chan
arXiv 2021. [Paper]
07 Jun 2021

A Variational Perspective on Diffusion-Based Generative Models and Score Matching
Chin-Wei Huang, Jae Hyun Lim, Aaron Courville
ICML Workshop 2021. [Paper] [Github]
5 Jun 2021

On Fast Sampling of Diffusion Probabilistic Models
Zhifeng Kong, Wei Ping
ICML Workshop 2021. [Paper] [Github]
31 May 2021

Cascaded Diffusion Models for High Fidelity Image Generation
Jonathan Ho1, Chitwan Saharia1, William Chan, David J. Fleet, Mohammad Norouzi, Tim Salimans
arXiv 2021. [Paper] [Project]
30 May 2021

Diffusion Models Beat GANs on Image Synthesis
Prafulla Dhariwal1, Alex Nichol1
arXiv 2021. [Paper] [Github]
11 May 2021

Image Super-Resolution via Iterative Refinement
Chitwan Saharia, Jonathan Ho, William Chan, Tim Salimans, David J. Fleet, Mohammad Norouzi
arXiv 2021. [Paper] [Project] [Github]
15 Apr 2021

Noise Estimation for Generative Diffusion Models
Robin San-Roman1, Eliya Nachmani1, Lior Wolf
arXiv 2021. [Paper]
6 Apr 2021

Diffusion Probabilistic Models for 3D Point Cloud Generation
Shitong Luo, Wei Hu
CVPR 2021. [Paper] [Github]
2 Mar 2021

Improved Denoising Diffusion Probabilistic Models
Alex Nichol1, Prafulla Dhariwal1
ICLR 2021. [Paper] [Github]
18 Feb 2021

Maximum Likelihood Training of Score-Based Diffusion Models
Yang Song1, Conor Durkan1, Iain Murray, Stefano Ermon
arXiv 2021. [Paper]
22 Jan 2021

Learning Energy-Based Models by Diffusion Recovery Likelihood
Ruiqi Gao, Yang Song, Ben Poole, Ying Nian Wu, Diederik P. Kingma
ICLR 2021. [Paper] [Github]
15 Dec 2020

Score-Based Generative Modeling through Stochastic Differential Equations
Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Kumar, Stefano Ermon, Ben Poole
ICLR 2021 (Oral). [Paper] [Github]
26 Nov 2020

Denoising Diffusion Implicit Models
Jiaming Song, Chenlin Meng, Stefano Ermon
ICLR 2021. [Paper] [Github]
6 Oct 2020

Denoising Diffusion Probabilistic Models
Jonathan Ho, Ajay Jain, Pieter Abbeel
NeurIPS 2020. [Paper] [Github] [Github]
19 Jun 2020

Improved Techniques for Training Score-Based Generative Models
Yang Song, Stefano Ermon
NeurIPS 2020. [Paper] [Github]
16 Jun 2020

Generative Modeling by Estimating Gradients of the Data Distribution
Yang Song, Stefano Ermon
NeurIPS 2019. [Paper] [Project] [Github]
12 Jul 2019

Neural Stochastic Differential Equations: Deep Latent Gaussian Models in the Diffusion Limit
Belinda Tzen, Maxim Raginsky
arXiv 2019. [Paper]
23 May 2019

Deep Unsupervised Learning using Nonequilibrium Thermodynamics
Jascha Sohl-Dickstein, Eric A. Weiss, Niru Maheswaranathan, Surya Ganguli
ICML 2015. [Paper] [Github]
2 Mar 2015

A Connection Between Score Matching and Denoising Autoencoders
Pascal Vincent
Neural Computation 2011. [Paper]
7 Jul 2011

Super Resolution

ILVR: Conditioning Method for Denoising Diffusion Probabilistic Models
Jooyoung Choi, Sungwon Kim, Yonghyun Jeong, Youngjune Gwon, Sungroh Yoon
ICCV 2021 (Oral). [Paper] [Github]
6 Aug 2021

Cascaded Diffusion Models for High Fidelity Image Generation
Jonathan Ho1, Chitwan Saharia1, William Chan, David J. Fleet, Mohammad Norouzi, Tim Salimans
arXiv 2021. [Paper] [Project]
30 May 2021

Image Super-Resolution via Iterative Refinement
Chitwan Saharia, Jonathan Ho, William Chan, Tim Salimans, David J. Fleet, Mohammad Norouzi
arXiv 2021. [Paper] [Project] [Github]
15 Apr 2021

Image-to-Image Translation

ILVR: Conditioning Method for Denoising Diffusion Probabilistic Models
Jooyoung Choi, Sungwon Kim, Yonghyun Jeong, Youngjune Gwon, Sungroh Yoon
ICCV 2021 (Oral). [Paper] [Github]
6 Aug 2021

UNIT-DDPM: UNpaired Image Translation with Denoising Diffusion Probabilistic Models
Hiroshi Sasaki, Chris G. Willcocks, Toby P. Breckon
arXiv 2021. [Paper]
12 Apr 2021

Image Inpainting

ILVR: Conditioning Method for Denoising Diffusion Probabilistic Models
Jooyoung Choi, Sungwon Kim, Yonghyun Jeong, Youngjune Gwon, Sungroh Yoon
ICCV 2021 (Oral). [Paper] [Github]
6 Aug 2021

SDEdit: Image Synthesis and Editing with Stochastic Differential Equations
Chenlin Meng, Yang Song, Jiaming Song, Jiajun Wu, Jun-Yan Zhu, Stefano Ermon
arXiv 2021. [Paper] [Project] [Github]
2 Aug 2021

Audio Generation

Variational Diffusion Models
Diederik P. Kingma, Tim Salimans, Ben Poole, Jonathan Ho
arXiv 2021. [Paper] [Github]
1 Jul 2021

PriorGrad: Improving Conditional Denoising Diffusion Models with Data-Driven Adaptive Prior
Sang-gil Lee, Heeseung Kim, Chaehun Shin, Xu Tan, Chang Liu, Qi Meng, Tao Qin, Wei Chen, Sungroh Yoon, Tie-Yan Liu
arXiv 2021. [Paper] [Project]
11 Jun 2021

Symbolic Music Generation with Diffusion Models
Gautam Mittal, Jesse Engel, Curtis Hawthorne, Ian Simon
arXiv 2021. [Paper] [Code] 30 Mar 2021

DiffWave with Continuous-time Variational Diffusion Models
Zhifeng Kong, Wei Ping, Jiaji Huang, Kexin Zhao, Bryan Catanzaro
ICLR 2021 [Paper] [Project] [Github] 21 Sep 2020

DiffWave: A Versatile Diffusion Model for Audio Synthesis
Jascha Sohl-Dickstein, Eric A. Weiss, Niru Maheswaranathan, Surya Ganguli
ICML 2021 (Oral) [Paper] [Github] [Github2]
21 Sep 2020

WaveGrad: Estimating Gradients for Waveform Generation
Nanxin Chen, Yu Zhang, Heiga Zen, Ron J. Weiss, Mohammad Norouzi, William Chan
ICLR 2021. [Paper] [Project] [Github]
2 Sep 2020

Audio Enhancement

A Study on Speech Enhancement Based on Diffusion Probabilistic Model
Yen-Ju Lu1, Yu Tsao1, Shinji Watanabe
arXiv 2021. [Paper]
25 Jul 2021

Restoring degraded speech via a modified diffusion model
Jianwei Zhang, Suren Jayasuriya, Visar Berisha
Interspeech 2021. [Paper]
22 Apr 2021

NU-Wave: A Diffusion Probabilistic Model for Neural Audio Upsampling
Junhyeok Lee, Seungu Han
Interspeech 2021. [Paper] [Project] [Github]
6 Apr 2021

Audio Conversion

DiffSVC: A Diffusion Probabilistic Model for Singing Voice Conversion
Songxiang Liu1, Yuewen Cao1, Dan Su, Helen Meng
arXiv 2021. [Paper] [Github]
28 May 2021

Text-to-Speech

Diff-TTS: A Denoising Diffusion Model for Text-to-Speech
Myeonghun Jeong, Hyeongju Kim, Sung Jun Cheon, Byoung Jin Choi, Nam Soo Kim
Interspeech 2021. [Paper]
3 Apr 2021

Time-series Forecasting

Autoregressive Denoising Diffusion Models for Multivariate Probabilistic Time Series Forecasting
Kashif Rasul, Calvin Seward, Ingmar Schuster, Roland Vollgraf
ICLR 2021. [Paper]
2 Feb 2021

Data Imputation

CSDI: Conditional Score-based Diffusion Models for Probabilistic Time Series Imputation
Yusuke Tashiro, Jiaming Song, Yang Song, Stefano Ermon
arXiv 2021. [Paper]
7 Jul 2021

Handwriting Synthesis

Diffusion models for Handwriting Generation
Troy Luhman1, Eric Luhman1
arXiv 2020. [Paper] [Github]
13 Nov 2020