---
title: "CoRe^2: Collect, Reflect and Refine to Generate Better and Faster"
canonical_url: "https://www.modelscope.cn/papers/126707"
md_url: "https://www.modelscope.cn/papers/126707.md"
arxiv_id: 2503.09662
published: 2025-03-12
last_updated: 2025-03-12
authors:
  - "Shitong Shao"
  - "Zikai Zhou"
  - "Dian Xie"
  - "Yuetong Fang"
  - "Tian Ye"
  - "Lichen Bai"
  - "Zeke Xie"
model_name: CoRe2
model_developer: "香港科技大学（广州）"
domain:
  - "自然语言处理"
  - "计算机视觉"
  - "深度学习"
type:
  - "自然语言处理"
  - "计算机视觉"
  - "深度学习"
  - "Computer Vision and Pattern Recognition (cs.CV)"
arxiv_url: "https://arxiv.org/abs/2503.09662"
pdf_url: "https://arxiv.org/pdf/2503.09662.pdf"
---

# CoRe^2: Collect, Reflect and Refine to Generate Better and Faster

> Making text-to-image (T2I) generative model sample both fast and well represents a promising research direction. Previous studies have typically focused on either enhancing the visual quality of synthesized images at the expense of sampling efficiency or…

「CoRe^2: Collect, Reflect and Refine to Generate Better and Faster」是 ModelScope 魔搭社区收录的论文，arXiv 2503.09662，作者为 Shitong Shao, Zikai Zhou, Dian Xie et al.，发表于 2025-03-12，属于 自然语言处理、计算机视觉、深度学习 领域。

- **ArXiv**: 2503.09662
- **Published**: 2025-03-12
- **Authors**: Shitong Shao, Zikai Zhou, Dian Xie, Yuetong Fang, Tian Ye, Lichen Bai, Zeke Xie
- **Model**: CoRe2
- **Developer**: 香港科技大学（广州）
- **Domain**: 自然语言处理, 计算机视觉, 深度学习
- **ArXiv URL**: https://arxiv.org/abs/2503.09662
- **PDF**: https://arxiv.org/pdf/2503.09662.pdf

Source: https://www.modelscope.cn/papers/126707

---

> CoRe2：快速高效生成高质量图像的新型即插即用框架

## 摘要

本文的研究背景是文本到图像生成（T2I）模型的采样效率与质量之间的权衡问题。传统方法要么以牺牲采样效率为代价提升视觉质量，要么加速采样但未能改善基础模型的生成能力，且大多数推理方法在扩散模型（DMs）和自回归模型（ARMs）上表现不稳定。为此，本文提出了一种新颖的即插即用推理范式CoRe2，包含三个子过程：收集（Collect）、反射（Reflect）和精炼（Refine）。CoRe2首先通过收集无分类器引导（CFG）轨迹来构建数据集，在反射阶段训练一个轻量级模型以反映易学内容并减少推理期间的功能评估次数，最后利用弱到强引导（W2S）精炼条件输出，增强模型生成高频和真实内容的能力。实验表明，CoRe2在多个基准测试中表现出显著性能改进，如HPD v2、Pick-of-Pic等，并能无缝集成最先进的Z-Sampling算法，同时节省时间。此外，理论分析进一步解释了CoRe2为何能有效提升推理性能。

## Abstract

Making text-to-image (T2I) generative model sample both fast and well represents a promising research direction. Previous studies have typically focused on either enhancing the visual quality of synthesized images at the expense of sampling efficiency or dramatically accelerating sampling without improving the base model's generative capacity. Moreover, nearly all inference methods have not been able to ensure stable performance simultaneously on both diffusion models (DMs) and visual autoregressive models (ARMs). In this paper, we introduce a novel plug-and-play inference paradigm, CoRe^2, which comprises three subprocesses: Collect, Reflect, and Refine. CoRe^2 first collects classifier-free guidance (CFG) trajectories, and then use collected data to train a weak model that reflects the easy-to-learn contents while reducing number of function evaluations during inference by half. Subsequently, CoRe^2 employs weak-to-strong guidance to refine the conditional output, thereby improving the model's capacity to generate high-frequency and realistic content, which is difficult for the base model to capture. To the best of our knowledge, CoRe^2 is the first to demonstrate both efficiency and effectiveness across a wide range of DMs, including SDXL, SD3.5, and FLUX, as well as ARMs like LlamaGen. It has exhibited significant performance improvements on HPD v2, Pick-of-Pic, Drawbench, GenEval, and T2I-Compbench. Furthermore, CoRe^2 can be seamlessly integrated with the state-of-the-art Z-Sampling, outperforming it by 0.3 and 0.16 on PickScore and AES, while achieving 5.64s time saving using this http URL is released at this https URL.