UniSHARP: Universal Sharp Monocular View Synthesis Across Camera Systems

Forum topic · 小凯 · 2026-06-09

Summary

UniSHARP (arXiv 2506.08646) extends SHARP, a popular photorealistic view synthesis method, to universal monocular rendering across a continuum of camera systems, from conventional perspective cameras to wide-field-of-view, fisheye, and omnidirectional panoramic settings. To overcome SHARP's pinhole-specific assumptions, the key idea is to align various images in a unified omnidirectional latent space. UniSHARP performs implicit alignment in both feature and Gaussian spaces: Gaussian primitives are arranged along rays and radial distances in a ray-based universal representation, while 2D semantic and 3D spatial features extracted from UniK3D-inspired encoders are jointly decoded to generate the complete Gaussian cloud. The authors also construct a benchmark covering diverse scenes and imaging systems, stratified by field of view (FoV) for fine-grained evaluation of universal monocular rendering. Extensive experiments show UniSHARP significantly outperforms existing methods.

Introduction

This paper proposes UniSHARP, an extension of SHARP — a popular photorealistic view synthesis method — for universal monocular rendering across a continuum of camera systems, from conventional perspective cameras to wide-field-of-view, fisheye, and omnidirectional panoramic settings.

Key Idea

To overcome the pinhole-specific assumptions of SHARP, the core idea is to align various images in a unified omnidirectional latent space. UniSHARP performs implicit alignment in both feature space and Gaussian space:

Gaussian primitives are arranged along rays and radial distances in a ray-based universal representation.
2D semantic and 3D spatial features extracted from UniK3D-inspired encoders are jointly decoded to generate the complete Gaussian cloud.

Evaluation

To comprehensively assess performance, the authors build a benchmark covering diverse scenes and imaging systems, stratified by field of view (FoV) to enable fine-grained evaluation of universal monocular rendering. Extensive experiments demonstrate that UniSHARP significantly outperforms existing methods.

Paper Info

Field: Computer Vision
Authors: Meixi Song, Dizhe Zhang, Hao Ren
Published: 2025-06-11
arXiv: 2506.08646