MEt3R: Measuring Multi-View Consistency in Generated Images

TL;DR

The paper introduces MEt3R, a new metric for evaluating multi-view consistency in generated images, enabling independent assessment of generated views without relying on scene-specific details.

Contribution

It proposes MEt3R, a novel, sampling-independent metric for measuring multi-view consistency in generated images, utilizing dense 3D reconstructions and feature map comparisons.

Findings

MEt3R effectively evaluates multi-view consistency across various generative models.

The metric reveals differences in consistency quality among existing methods.

It provides a new benchmark for assessing multi-view image generation quality.

Abstract

We introduce MEt3R, a metric for multi-view consistency in generated images. Large-scale generative models for multi-view image generation are rapidly advancing the field of 3D inference from sparse observations. However, due to the nature of generative modeling, traditional reconstruction metrics are not suitable to measure the quality of generated outputs and metrics that are independent of the sampling procedure are desperately needed. In this work, we specifically address the aspect of consistency between generated multi-view images, which can be evaluated independently of the specific scene. Our approach uses DUSt3R to obtain dense 3D reconstructions from image pairs in a feed-forward manner, which are used to warp image contents from one view into the other. Then, feature maps of these images are compared to obtain a similarity score that is invariant to view-dependent effects. Using MEt3R, we evaluate the consistency of a large set of previous methods for novel view and video generation, including our open, multi-view latent diffusion model.