Diffusing Differentiable Representations

TL;DR

This paper presents a training-free technique for sampling differentiable representations using pretrained diffusion models, improving the quality and diversity of generated images, panoramas, and 3D objects by leveraging a novel reverse-time process.

Contribution

It introduces a new method that pulls back the diffusion process to sample differentiable representations without additional training, enhancing consistency and detail.

Findings

Significantly improved quality of diffreps for images and 3D objects.

Enhanced diversity in generated samples across multiple domains.

Demonstrated general applicability of the method to various data types.

Abstract

We introduce a novel, training-free method for sampling differentiable representations (diffreps) using pretrained diffusion models. Rather than merely mode-seeking, our method achieves sampling by "pulling back" the dynamics of the reverse-time process--from the image space to the diffrep parameter space--and updating the parameters according to this pulled-back process. We identify an implicit constraint on the samples induced by the diffrep and demonstrate that addressing this constraint significantly improves the consistency and detail of the generated objects. Our method yields diffreps with substantially improved quality and diversity for images, panoramas, and 3D NeRFs compared to existing techniques. Our approach is a general-purpose method for sampling diffreps, expanding the scope of problems that diffusion models can tackle.