RORD-50 Dataset

This repository contains the RORD-50 dataset, introduced in the paper From Ideal to Real: Stable Video Object Removal under Imperfect Conditions.

Project Page | GitHub | Paper

Introduction

The RORD-50 dataset is a benchmark designed to evaluate video object removal performance under real-world challenges, such as shadows, abrupt motion, and defective masks. It was introduced as part of the Stable Video Object Removal (SVOR) framework, which focuses on achieving shadow-free, flicker-free, and mask-defect-tolerant removal.

Overview

Removing objects from videos remains difficult in the presence of real-world imperfections. SVOR advances video object removal from ideal settings toward real-world applications by handling abrupt motion and mask defects effectively. This dataset provides the necessary benchmarks for testing the robustness and temporal stability of video inpainting models.

Citation

If you find this dataset or the SVOR framework useful for your research, please consider citing the paper:

@article{hu2026svor,
   title={From Ideal to Real: Stable Video Object Removal under Imperfect Conditions},
   author={Hu, Jiagao and Chen, Yuxuan and Li, Fuhao and Wang, Zepeng and Wang, Fei and Daiguo, Zhou and Luan, Jian},
   journal={arXiv preprint arXiv:2603.09283},
   year={2026}
}

Acknowledgement

This work benefits from the following open-source projects:

• VideoX-Fun
• VACE
• ROSE
• SAM2 - Segment Anything Model 2
• RORD