Reflect3r
Single-View 3D Stereo Reconstruction Aided by Mirror Reflections
3DV 2026
Jing Wu Zirui Wang Iro Laina Victor Adrian Prisacariu
University of Oxford
Citation
@article{wu2026reflect3r,
author = {Wu, Jing and Wang, Zirui and Laina, Iro and Prisacariu, Victor},
title = {{Reflect3r: Single-View 3D Stereo Reconstruction Aided by Mirror Reflections}},
journal = {3DV},
year = {2026},
}
Abstract
Mirror reflections are common in everyday environments and can provide stereo information within a single capture, as the real and reflected virtual views are visible simultaneously. We exploit this property by treating the reflection as an auxiliary view and designing a transformation that constructs a physically valid virtual camera, allowing direct pixel-domain generation of the virtual view while adhering to the real-world imaging process. This enables a multi-view stereo setup from a single image, simplifying the imaging process, making it compatible with powerful feed-forward reconstruction models for generalizable and robust 3D reconstruction. To further exploit the geometric symmetry introduced by mirrors, we propose a symmetric-aware loss to refine pose estimation. Our framework also naturally extends to dynamic scenes, where each frame contains a mirror reflection, enabling efficient per-frame geometry recovery. For quantitative evaluation, we provide a fully customizable synthetic dataset of 16 Blender scenes, each with ground-truth point clouds and camera poses. Extensive experiments on real-world data and synthetic data are conducted to illustrate the effectiveness of our method.
Method
Reflect3r reconstructs 3D scenes from a single-view image by leveraging mirror reflections. A reflection transformation is designed to ensure that flipping the real view in the pixel domain, simulating a virtual camera imaging, enables seamless integration with modern feed-forward models. Following the initial prediction, the reflection transformation is used as a geometric constraint to refine pose optimization.
Synthetic Blender Data 
Thumbnails of the dataset. where each image represents a fully customizable Blender scene.
Download
🤗 HuggingFace Data- Download the original editable Blender scenes (
*.blend) from here. - Download the rendered and processed data (ground-truth point clouds) from here.
- Check here and here for more details about our data.
Preview
We visualize some examples of the synthetic ground-truth point clouds below for preview.
Use the controls to switch between examples.
Results
We show the reconstruction results on the synthetic data.
Acknowledgements
Zirui Wang is supported by an ARIA research gift grant from Meta Reality Lab.
Our Code is built upon:
DUSt3R
The 16 Blender scenes are collected from online websites, we rearranged and cleaned the scenes and modelled a mirror on top of them for research purpose. Here we listed all the original download links for these scenes, we thank these designers for their great work.