论文阅读

必读

DDPM: [Denoising Diffusion Probabilistic Models]
- [Improved Denoising Diffusion Probabilistic Models] (ICML 2021)

DDIM: [Denoising Diffusion Implicit Models]
SDE: [Score-Based Generative Modeling through Stochastic Differential Equations]
- 前置： [Generative Modeling by Estimating Gradients of the Data Distribution]
LDM: [High-Resolution Image Synthesis with Latent Diffusion Models]

搞懂之后可以复现如何通过 Classifier-Guided 和 Classifier-Free 来控制生成
Classifier-Guided: [Diffusion Models Beat GANs on Image Synthesis]
Classifier-Free: [Classifier-Free Diffusion Guidance]
Imagen: [Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding]
[DALL-E 2]
[DALL-E 3]
搞懂后可以探索如何微调现有 diffusion model，对其注入特定的概念
DreamBooth: [DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation]
LoRA: [LoRA: Low-Rank Adaptation of Large Language Models]
Textual Inversion: [An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion]
再探究如何实现细粒度控制
ControlNet: [Adding Conditional Control to Text-to-Image Diffusion Models]
Transformer DiT: [Scalable Diffusion Models with Transformers]
[Inversion]: [Null-text Inversion for Editing Real Images using Guided Diffusion Models]
分类中的应用：
[Your Diffusion Model is Secretly a Zero-Shot Classifier]
[Identity Preserving]
https://github.com/xuekt98/readed-papers/tree/main
tutorials: https://geometry.cs.ucl.ac.uk/courses/diffusion4VC_eg24/