# UniPR: Unified Object-level Real-to-Sim Perception and Reconstruction from a Single Stereo Pair

Source: [UniPR: Unified Object-level Real-to-Sim Perception and Reconstruction from a Single Stereo Pair](<https://www.sota2.com/research/paper/69c0c8d60636e5bf4afc3937>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-24T05:00:03.763Z

Authors: Chuanrui Zhang, Yingshuang Zou, ZhengXian Wu, Yonggen Ling, Yuxiao Yang, Ziwei Wang

Published: 2026-03-20T03:47:48.000Z

## Abstract

Perceiving and reconstructing objects from images are critical for real-to-sim transfer tasks, which are widely used in the robotics community. Existing methods rely on multiple submodules such as detection, segmentation, shape reconstruction, and pose estimation to complete the pipeline. However, such modular pipelines suffer from inefficiency and cumulative error, as each stage operates on only partial or locally refined information while discarding global context. To address these limitations, we propose UniPR, the first end-to-end object-level real-to-sim perception and reconstruction framework. Operating directly on a single stereo image pair, UniPR leverages geometric constraints to resolve the scale ambiguity. We introduce Pose-Aware Shape Representation to eliminate the need for per-category canonical definitions and to bridge the gap between reconstruction and pose estimation tasks. Furthermore, we construct a large-vocabulary stereo dataset, LVS6D, comprising over 6,300 objects, to facilitate large-scale research in this area. Extensive experiments demonstrate that UniPR reconstructs all objects in a scene in parallel within a single forward pass, achieving significant efficiency gains and preserves true physical proportions across diverse object types, highlighting its potential for practical robotic applications.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Category-level 6D object pose estimation on LVS6D Easy](<https://www.sota2.com/research/sota/category-level-6d-object-pose-estimation-on-lvs6d-easy.md>): Category-level 6D Object Pose Estimation; LVS6D Easy; AP: 75.9.

- [Category-level 6D object pose estimation on LVS6D Medium](<https://www.sota2.com/research/sota/category-level-6d-object-pose-estimation-on-lvs6d-medium.md>): Category-level 6D Object Pose Estimation; LVS6D Medium; AP: 82.4.

- [Category-level 6D object pose estimation on LVS6D (Hard)](<https://www.sota2.com/research/sota/category-level-6d-object-pose-estimation-on-lvs6d-hard.md>): Category-level 6D Object Pose Estimation; LVS6D (Hard); AP: 75.2.

- [3D Reconstruction on LVS6D 50-object subset (test)](<https://www.sota2.com/research/sota/3d-reconstruction-on-lvs6d-50-object-subset-test.md>): 3D Reconstruction; LVS6D 50-object subset (test); CD: 0.0083.

- [6D Object Pose Estimation on SS3D](<https://www.sota2.com/research/sota/6d-object-pose-estimation-on-ss3d.md>): 6D Object Pose Estimation; SS3D; 3D50 Accuracy: 89.3.

- [6D Object Pose Estimation on TOD](<https://www.sota2.com/research/sota/6d-object-pose-estimation-on-tod.md>): 6D Object Pose Estimation; TOD; 10° 10cm Success Rate: 86.9.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2603.19616>)

[PDF](<https://arxiv.org/pdf/2603.19616.pdf>)