arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2406.09413cs.CVcs.GRcs.LG
收藏

定制化扩散模型权重空间的可解释性研究

Interpreting the Weight Space of Customized Diffusion Models

UC Berkeley(加州大学伯克利分校);Snap Inc.(Snap公司);Stanford University(斯坦福大学)

Amil Dravid, Yossi Gandelsman, Kuan-Chieh Wang, Rameen Abdal, Gordon Wetzstein, Alexei A. Efros, Kfir Aberman

AI总结:

本文通过构建超过60,000个定制化扩散模型的数据集,提出weights2weights权重子空间,并展示其在采样、编辑和反演中的应用,证明微调扩散模型的权重空间可作为可解释的元潜在空间生成新模型。

摘要:

我们研究由大量定制化扩散模型所张成的权重空间。我们通过构建一个包含超过60,000个模型的数据集来填充该空间,其中每个模型都是一个基础模型经过微调后用于插入不同人物的视觉身份。我们将这些权重的底层流形建模为一个子空间,并将其命名为weights2weights。我们展示了该空间的三个直接应用,它们能够产生新的扩散模型——采样、编辑和反演。首先,从该空间中采样一组权重会得到一个编码全新身份的新模型。其次,我们在该空间中找到了与身份语义编辑(例如,添加胡须)相对应的线性方向,从而得到一个对原始身份进行编辑后的新模型。最后,我们证明将单张图像反演到该空间可以将一个真实身份编码进模型中,即使输入图像是分布外样本(例如,一幅绘画)。我们进一步发现,扩散模型权重空间的这些线性性质可以扩展到其他视觉概念。我们的结果表明,微调后扩散模型的权重空间可以表现为一个可解释的元潜在空间,从而产生新的模型。

英文摘要:

We investigate the space of weights spanned by a large collection of customized diffusion models. We populate this space by creating a dataset of over 60,000 models, each of which is a base model fine-tuned to insert a different person's visual identity. We model the underlying manifold of these weights as a subspace, which we term weights2weights. We demonstrate three immediate applications of this space that result in new diffusion models -- sampling, editing, and inversion. First, sampling a set of weights from this space results in a new model encoding a novel identity. Next, we find linear directions in this space corresponding to semantic edits of the identity (e.g., adding a beard), resulting in a new model with the original identity edited. Finally, we show that inverting a single image into this space encodes a realistic identity into a model, even if the input image is out of distribution (e.g., a painting). We further find that these linear properties of the diffusion model weight space extend to other visual concepts. Our results indicate that the weight space of fine-tuned diffusion models can behave as an interpretable meta-latent space producing new models.

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。

↑