ArticleMetadataMain page

source · 2026-04-15

Specialist Diffusion（CVPR 2023）：可插拔且样本高效的文本到图像扩散微调

#near-cvpr-2025#image-generation#diffusion#efficiency#primary-source

Specialist Diffusion（CVPR 2023）：可插拔且样本高效的文本到图像扩散微调

会议：CVPR 2023
发表日期：2023-01-01
资料加入日期：2026-04-14

一句话结论

这篇工作把 sample-efficient fine-tuning 路线补进库里，强化了文本到图像扩散模型的轻量适配分支。

问题定义

它要解决的是文本到图像扩散模型学习新风格或新概念时通常需要较高数据与调参成本的问题。对当前知识库来说，它补的是 DreamBooth / personalization 之外更插件式的适配路线。

方法概述

Specialist Diffusion 通过 plug-and-play、sample-efficient fine-tuning 方式，让 text-to-image diffusion models 能用较少样本学会未见风格或概念。

关键发现

  • 它说明 image-generation 的适配路线不只靠全量微调，也在追求更轻量、更可插拔的 specialist 模块。
  • 它让 diffusion-models 页对参数高效/样本高效适配的覆盖更完整。
  • 它适合以后做“低资源改进点”方向探索。

局限或疑问

  • 轻量适配通常伴随能力边界限制。
  • 它更偏风格/概念适配，不直接代表通用编辑能力。
  • 它是高价值适配路线，但仍需和更重微调路线一起比较。

原始链接

  • https://openaccess.thecvf.com/content/CVPR2023/html/Lu_Specialist_Diffusion_Plug-and-Play_Sample-Efficient_Fine-Tuning_of_Text-to-Image_Diffusion_Models_To_CVPR_2023_paper.html
  • https://openaccess.thecvf.com/content/CVPR2023/papers/Lu_Specialist_Diffusion_Plug-and-Play_Sample-Efficient_Fine-Tuning_of_Text-to-Image_Diffusion_Models_To_CVPR_2023_paper.pdf

相关页面

备注

Specialist Diffusion 在这套库里的作用，是把样本高效的扩散适配路线补成明确节点。

Metadata

{
  "id": "2026-04-14-specialist-diffusion",
  "type": "source",
  "title": "Specialist Diffusion（CVPR 2023）：可插拔且样本高效的文本到图像扩散微调",
  "status": "reviewed",
  "created": "2026-04-14",
  "updated": "2026-04-15",
  "venue": "CVPR 2023",
  "ingested_at": "2026-04-14",
  "tags": [
    "near-cvpr-2025",
    "image-generation",
    "diffusion",
    "efficiency",
    "primary-source"
  ],
  "note_status": "reviewed",
  "source_type": "paper",
  "authors": [
    "Haoming Lu",
    "Hazarapet Tunanyan",
    "Kai Wang",
    "Shant Navasardyan",
    "Zhangyang Wang",
    "Humphrey Shi"
  ],
  "published_at": "2023-01-01",
  "canonical_links": [
    "https://openaccess.thecvf.com/content/CVPR2023/html/Lu_Specialist_Diffusion_Plug-and-Play_Sample-Efficient_Fine-Tuning_of_Text-to-Image_Diffusion_Models_To_CVPR_2023_paper.html",
    "https://openaccess.thecvf.com/content/CVPR2023/papers/Lu_Specialist_Diffusion_Plug-and-Play_Sample-Efficient_Fine-Tuning_of_Text-to-Image_Diffusion_Models_To_CVPR_2023_paper.pdf"
  ],
  "raw_entry": "raw/ingest/2026-04-14-specialist-diffusion/",
  "topics": [
    "topics/image-generation",
    "topics/diffusion-models"
  ],
  "entities": [
    "entities/unified-image-generation-editing"
  ],
  "claims": [],
  "questions": [
    "questions/question-will-unified-image-models-sustain-their-advantage"
  ]
}