Summary
OmniRay is a feed-forward monocular 3D reconstruction model that works across diverse camera geometries, including fisheye and omnidirectional lenses, without requiring camera information at inference time. Most monocular models assume a pinhole camera and produce distorted geometry on wide-angle imagery, while camera-aware models predict per-pixel ray fields but rely on a single fixed basis. OmniRay introduces a Ray Module with three hierarchical components: a monotone radial base capturing radially symmetric structure, a spherical-harmonic residual representing smooth deviations, and a free-form residual for remaining local variations, composed as successive rotations on the sphere. Given a single image, the model predicts a ray direction and an up-to-scale radial distance per pixel, from which 3D point clouds are recovered. Experiments across camera configurations establish a new state of the art in camera-agnostic monocular 3D reconstruction.
Paper Overview
Field: Computer Vision
Authors: Namitha Guruprasad, Abhay Yadav, Cheng Peng, Rama Chellappa, Anand Bhattad
Published: 2026-10-22
arXiv: 2610.31645
Abstract
Monocular models recover 3D geometry from a single image in one pass, but most assume a pinhole camera and produce distorted geometry on wide-angle imagery. Camera-aware models remove this assumption by predicting a per-pixel ray field, but they represent different camera geometries using a single fixed basis.
We argue that camera geometry has a hierarchical complexity. OmniRay's feed-forward Ray Module consists of three components:
- A monotone radial base that captures radially symmetric structure
- A spherical-harmonic residual that represents smooth deviations
- A free-form residual that captures remaining local variations
These components are composed as successive rotations on the sphere, allowing the representation to accommodate diverse camera geometries.
Given a single image, OmniRay predicts a ray direction and an up-to-scale radial distance for every pixel, from which a 3D point cloud is recovered — with no camera information required at inference time. Extensive experiments across different camera configurations show that OmniRay sets a new state of the art in camera-agnostic monocular 3D reconstruction.
---
*Auto-collected on 2026-10-24*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178636412