StereoSpace: Depth-Free Synthesis of Stereo Geometry via End-to-End Diffusion in a Canonical Space

TL;DR

StereoSpace is a diffusion-based framework that synthesizes stereo images from monocular inputs without explicit depth, using viewpoint conditioning and a canonical space, achieving high-quality results and robust evaluation.

Contribution

It introduces a depth-free, viewpoint-conditioned diffusion approach for stereo synthesis with a new evaluation protocol and state-of-the-art performance.

Findings

Outperforms existing warp & inpaint, latent-warping, and warped-conditioning methods.

Achieves sharp parallax and robustness on layered and non-Lambertian scenes.

Establishes viewpoint-conditioned diffusion as a scalable stereo generation solution.

Abstract

We introduce StereoSpace, a diffusion-based framework for monocular-to-stereo synthesis that models geometry purely through viewpoint conditioning, without explicit depth or warping. A canonical rectified space and the conditioning guide the generator to infer correspondences and fill disocclusions end-to-end. To ensure fair and leakage-free evaluation, we introduce an end-to-end protocol that excludes any ground truth or proxy geometry estimates at test time. The protocol emphasizes metrics reflecting downstream relevance: iSQoE for perceptual comfort and MEt3R for geometric consistency. StereoSpace surpasses other methods from the warp & inpaint, latent-warping, and warped-conditioning categories, achieving sharp parallax and strong robustness on layered and non-Lambertian scenes. This establishes viewpoint-conditioned diffusion as a scalable, depth-free solution for stereo generation.