Description
Generating 3D scenes from one photo leaves occluded parts missing and objects misplaced. 3D-Fixer from Horizon Robotics, CVPR 2026, proposes coarse-to-fine in-place completion of 3D scenes from a single image.
It ships the paper, model, dataset and an online demo for digital twins and more.
In-place:Objects stay put.
Coarse-to-fine:Details refined.
Single image:One photo.
Open:Model and dataset.
It ships the paper, model, dataset and an online demo for digital twins and more.
Features
In-place:Objects stay put.
Coarse-to-fine:Details refined.
Single image:One photo.
Open:Model and dataset.
Screenshots
Tags:3d-modeling