Generating 3D-consistent novel views from 2D diffusion models remains a fundamental challenge in computer vision. A new paper on arXiv introduces a framework called ESC that aims to address this by enhancing geometric alignment for monocular view synthesis.

The core idea is to escape the ambiguity inherent in 2D diffusion models by incorporating geometric constraints. This is intended to improve consistency when synthesizing new views from a single image, particularly under drastic changes.

The abstract is truncated in the announcement, so the precise form of the geometric constraints is not fully detailed. Still, the stated goal is clear: use geometry to guide diffusion and reduce the ambiguity that leads to inconsistent views.