Welcome to OccluScene3D Benchmark

Unleash the power of 3D large scene reconstruction and rendering with our comprehensive benchmark designed to evaluate algorithms in the presence of occlusion challenges. Whether you're a researcher, developer, or enthusiast, OccluScene3D offers the datasets you need to advance your work.

Get Started

Introduction

We present a comprehensive benchmark dedicated to image-based 3D reconstruction, which serves as a crucial benchmark for researchers and developers in this area. The benchmark sequences are carefully captured in realistic conditions, reflecting real-world complexity. We use a common mobile phone in wide - angle mode and landscape orientation to record these sequences at 60 Hz. This frame rate ensures a smooth and detailed visual stream, crucial for accurate 3D reconstruction. We extract 3% of the frames uniformly from each video sequence. Then, we use COLMAP to estimate the camera's intrinsic and extrinsic parameters. These parameters are essential for defining the spatial and geometric characteristics of the captured scenes. They form the basis for subsequent 3D reconstruction tasks, enabling precise and reliable development and evaluation of various algorithms.

Gallery

Canteen

Classbuilding A

Hall

Classbuilding B

Classbuilding C

Study Room

Parking Lot

Meeting Room

Paper

Supplementary

BibTeX

Please cite our work if you use data from this site.

@article{liu2025occlugaussian,
    title={OccluGaussian: Occlusion-Aware Gaussian Splatting for Large Scene Reconstruction and Rendering},
    author={Liu, Shiyong and Tang, Xiao and Li, Zhihao and He, Yingfan and Ye, Chongjie and Liu, Jianzhuang and Huang, Binxiao and Zhou, Shunbo and Wu, Xiaofei},
    journal={ICCV},
    year={2025}
}