[图像编辑综述01]

Diffusion Model-Based Image Editing: A Survey Link to Diffusion Model-Based Image Editing: A Survey

From：TPAMI 2025

前言 Link to 前言

作者认为，基于扩散模型的图像编辑方法，可以从3个角度去分类

• Learning strategies

  

• Input conditions

  

• Editing tasks

  

一、Training-based Approches Link to 一、Training-based Approches

• Domain-Specific Editing

  • CLIP Guidance

  • Cycling Regularization

  • Projection and Interpolation

    Hierarchical diffusion autoencoders and disentangled image manipulation

  • Classifier Guidance

• Reference and Attribute Guided Editing

  • Reference-Based Image Composition

• Instructional Editing

  • InstructPix2Pix Framework
  • Model Architecture Enhancement
  • Data Quality Enhancement
  • Human Feedback-Enhanced Learning
  • Visual Instruction
  • Leveraging Multimodal Large-Scale Models

• Pseudo-Target Retrieval Based Editing

二、Testing-time Finetuing Approaches Link to 二、Testing-time Finetuing Approaches

• Denoising Model Finetuning
  • Finetuning Entire Denoising Models
  • Partial Parameter Finetuning in Denoising Models
• Embedding Finetuning
  • Null-Text Embedding Finetuning
  • Text Embedding Finetuning
• Guidance with a Hypernetwork
• Latent Variable Optimization
  • Human-Guided Optimization of Latent Variables
  • Utilizing Network Layers and Input for Optimizing Latent Variables
• Hybrid Finetuning
  • Text Embedding & Denoising Model Finetuning
  • Text Encoder & Denoising Model Finetuning

三、Training and Fintuning Free Approaches Link to 三、Training and Fintuning Free Approaches

• Input Text Refinement
  • Enhancement of Input Text Embeddings
  • Instruction-Based Textual Guidance
• Inversion/Sampling Modification
  • Reconstruction Information Memory
  • Utilising Null-Text in Sampling
  • Single-Step Multiple Noise Prediction
• Attention Modification
  • Attention Map Replacement
  • Attention Feature Replacement
  • Local Attention Map Modification
  • Attention Score Guidance
• Mask Guidance
  • Mask-Enhanced Denoising Efficiency
  • Mask Auto-Generation
  • Mask-Guided Regional Focus
• Multi-Noise Redirection
  • Semantic Noise Steering
  • Object-Aware Noise Redirection