arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.08064cs.CV
收藏

通过一维代码分布评估生成模型

Evaluating Generative Models via One-Dimensional Code Distributions

WeChat AI, Tencent Inc.(腾讯微信AI);School of Intelligence Science and Technology, Peking University(北京大学智能科学与技术学院);College of Computer Science and Artificial Intelligence, Fudan University(复旦大学计算机科学与人工智能学院)

Zexi Jia, Pengcheng Luo, Yijia Zhong, Jinchao Zhang, Jie Zhou

AI总结:

本文提出基于离散视觉令牌空间的生成模型评估方法,引入CHD和CMMS两个度量标准,并构建VisForm基准以验证其在不同数据集上的有效性。

摘要:

大多数生成模型的评估依赖于特征分布度量标准,如FID,这些标准在连续识别特征上操作,这些特征被显式训练以对表现变化具有不变性,因此会丢弃对感知质量至关重要的线索。我们相反在离散视觉令牌的空间中评估模型,其中现代1D图像令牌化器能够紧凑地编码语义和感知信息,质量表现为可预测的令牌统计。我们引入了Codebook Histogram Distance (CHD),一种无需训练的令牌空间分布度量标准,以及Code Mixture Model Score (CMMS),一种从令牌序列的合成退化中学习的无参考质量度量标准。为了在广泛的分布转移下测试度量标准,我们进一步提出了VisForm,一个包含210万张图像、62种视觉形式和12种生成模型的基准,配有专家注释。在AGIQA、HPDv2/3和VisForm上,我们的基于令牌的度量标准在与人类判断的相关性上达到了最先进的水平。我们将发布所有代码和数据集以促进未来研究,代码已公开在https://github.com/zexiJia/1d-Distance。

英文摘要:

Most evaluations of generative models rely on feature-distribution metrics such as FID, which operate on continuous recognition features that are explicitly trained to be invariant to appearance variations, and thus discard cues critical for perceptual quality. We instead evaluate models in the space of discrete visual tokens, where modern 1D image tokenizers compactly encode both semantic and perceptual information and quality manifests as predictable token statistics. We introduce Codebook Histogram Distance (CHD), a training-free distribution metric in token space, and Code Mixture Model Score (CMMS), a no-reference quality metric learned from synthetic degradations of token sequences. To stress-test metrics under broad distribution shifts, we further propose VisForm, a benchmark of 210K images spanning 62 visual forms and 12 generative models with expert annotations. Across AGIQA, HPDv2/3, and VisForm, our token-based metrics achieve state-of-the-art correlation with human judgments. We will release all code and datasets to facilitate future research, with the code publicly available at https://github.com/zexiJia/1d-Distance.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑