オムライスの備忘録

数学・統計学・機械学習・プログラミングに関することを記す

トップ > データサイエンス > 【深層学習】Diffusion Model #まとめ編 #00

【深層学習】Diffusion Model #まとめ編 #00

データサイエンス データサイエンス-深層学習

Index

基本アルゴリズム

SBM と DDPM を合わせたモデルを拡散モデルと呼ぶ.

スコアベースモデル / SBM

Diffusion Probabilistic Model / DPM / 2015

Denoising Diffusion Probabilistic Model / DDPM / 2020

応用アルゴリズム

Latent Diffusion Model / LDM / 2021

Stable Diffusion v1

Guided Diffusion / 2021

画像生成の改善手法.

Diffusion Transformer / DiT 2022

Transformer の導入.

Imagen / 2022

eDiff-I / 2022

DPM-Solver / 2022

推論回数を減らすために、微分方程式を利用した手法.



  • DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps

EDM / 2022

拡散モデルを訓練・ネットワーク・サンプリングの3モジュールに分解して設計を再考した手法.



  • Elucidating the Design Space of Diffusion-Based Generative Models

Cold Diffusion / 2022

SceneDiffuser / 2023

  • Diffusion-based Generation, Optimization, and Planning in 3D Scenes

DIffuson-based Residual Augmentation Codec / DIRAC / 2023

  • Neural Image Compression with a Diffusion-Based Decoderv

Simple Diffusion / 2023

高解像度画像の生成の高速化.

  • simple diffusion: End-to-end diffusion for high resolution images

Mixture of Diffusers / 2023

  • Mixture of Diffusers for scene composition and high resolution image generation

Dual-CycleDiffusion / 2023

  • Eliminating Prior Bias for Semantic Image Editing via Dual-Cycle Diffusion

Design Booster / 2023

  • Design Booster: A Text-Guided Diffusion Model for Image Translation with Spatial Layout Preservation

SE3 / 2023

  • SE(3) diffusion model with application to protein backbone generation

Iterative Coherent Identity Injection / 2023

  • Zero-shot Generation of Coherent Storybook from Plain Text Story using Diffusion Models

UniPC / 2023

  • UniPC: A Unified Predictor-Corrector Framework for Fast Sampling of Diffusion Models

Q-Diffusion / 2023

  • Q-Diffusion: Quantizing Diffusion Models

PFGM++ / 2023

  • PFGM++: Unlocking the Potential of Physics-Inspired Generative Models

SR3+ / 2023

  • Denoising Diffusion Probabilistic Models for Robust Image Super-Resolution in the Wild

Projected latent Video Diffusion Model / PVDM / 2023

  • Video Probabilistic Diffusion Models in Projected Latent Space

Denoising Diffusion Operators / DDO / 2023

  • Score-based Diffusion Models in Function Space

Universal Guidance Diffusion / 2023

I2SB / 2023

  • I2SB: Image-to-Image Schrödinger Bridge

Single Motion Diffusion / 2023

PRedItOR / 2023

  • PRedItOR: Text Guided Image Editing with Diffusion Prior

MultiDiffusion / 2023

  • MultiDiffusion: Fusing Diffusion Paths for Controlled Image Generation

Latent Diffusion Prior / 2023

Composer / 2023

  • Composer: Creative and Controllable Image Synthesis with Composable Conditions

Cross-domain Compositing / 2023

  • Cross-domain Compositing with Pretrained Diffusion Models

Dual Pseudo Training / DPT / 2023

  • Diffusion Models and Semi-Supervised Learners Benefit Mutually with Few Labels

Differentially Private Diffusion Models / 2023

  • Differentially Private Diffusion Models Generate Useful Synthetic Images

Diffusion Probabilistic Fields / DPF / 2023

  • Diffusion Probabilistic Fields

TRACT / 2023

  • TRACT: Denoising Diffusion Models with Transitive Closure Time-Distillation

Cones / 2023

2 つの概念を合成した画像を生成.

  • Cones: Concept Neurons in Diffusion Models for Customized Generation

FGDS / 2023

  • Fast Diffusion Sampler for Inverse Problems by Geometric Decomposition

Min-SNR-γ / 2023

高速化.

  • Efficient Diffusion Training via Min-SNR Weighting Strategy

FreeDoM / 2023

学習済みモデルを利用する.

  • FreeDoM: Training-Free Energy-Guided Conditional Diffusion Model

WaveDiff / 2023

DOODL / 2023

  • End-to-End Diffusion Latent Optimization Improves Classifier Guidance

DiffCollage / 2023

Consistency Models

Patch Diffusion / 2023

  • Patch Diffusion: Faster and More Data-Efficient Training of Diffusion Models

タスク・データ分野

画像

3D

時系列

自然言語

Diffusion-LM / 2022

文章生成のタスク.



音響

動画像

マルチモーダル

Text-to-Image

Image Editing

テクニック・工夫

Model Editing

学習済みモデルを都合が良いように修正する.

  • Erasing Concepts from Diffusion Models

Watermarking

著作権などの対策.

  • A Recipe for Watermarking Diffusion Models

ReVersion / 2023

MAE

Diff MAE / 2023

Fine Turning

DiffFit / 2023

  • DiffFit: Unlocking Transferability of Large Diffusion Models via Simple Parameter-Efficient Fine-Tuning

アプリケーション・サービス

X diffusion / Picasso Diffusion

Textual Inversion Pipeline for Stable Diffusion

Cool Japan Diffusion

研究

  • Understanding the Diffusion Objective as a Weighted Integral of ELBOs

  • Diffusion Models are Minimax Optimal Distribution Estimators

拡散モデルは分布推定の意味でミニマックス最適な推定誤差を達成可能であることを示した.

分布のサポートが低次元である場合は次元の呪いを回避し，Wasserstein距離の意味で最適レートを達成することも示している．

参考

  • Deep Unsupervised Learning using Nonequilibrium Thermodynamics

    • [2015]
    • v8
    • Abstruct
    • 1 Introduction
      • 1.1 Diffusion probabilistic models
      • 1.2 Relationship to other work
    • 2 Algorithm
      • 2.1 Forward Trajectory
      • 2.2 Reverse Trajectory
      • 2.3 Model Probability
      • 2.4 Training
      • 2.5 Multiplying Distributions, and Computing Posteriors
      • 2.6 Entropy of Reverse Process
    • arxiv.org

  • GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models

    • [2021 OpenAI]
    • 2 Background
      • 2.1 Diffusion Models
    • arxiv.org

書籍

Web サイト

Tweet







動画