Long-RVOS: A Comprehensive Benchmark for Long-term Referring Video Object Segmentation

• GitHub Repository: https://github.com/iSEE-Laboratory/Long_RVOS
• Project Page: https://isee-laboratory.github.io/Long-RVOS/
• Paper: arXiv:2505.12702

Dataset Description

Dataset Summary

Long-RVOS is the first large-scale long-term referring video object segmentation benchmark, containing 2,000+ videos with an average duration exceeding 60 seconds. The dataset addresses the challenge of segmenting and tracking objects in long-form videos based on natural language descriptions, advancing the task towards more practical and realistic scenarios.

Dataset Statistics

• Total videos: 2,193
• Average video duration: 60.3 seconds
• Average frames per video: 361.7
• Object categories: 163
• Splits: Train, Validation, and Test sets

Dataset Structure

Data Organization

The dataset is organized as follows:

data/
└── long_rvos/
    ├── train/
    │   ├── JPEGImages/
    │   │   └── {video_id}/
    │   │       └── {frame_name}.jpg
    │   ├── Annotations/
    │   │   └── {video_id}/
    │   │       └── {object_id}/
    │   │           └── {frame_name}.png
    │   └── meta_expressions.json
    ├── valid/
    │   ├── JPEGImages/
    │   ├── Annotations/
    │   └── meta_expressions.json
    └── test/
        ├── JPEGImages/
        ├── Annotations/
        └── meta_expressions.json

Data Format

• JPEGImages: Video frames extracted and stored as JPEG images
• Annotations: Binary mask annotations (PNG format) for each object instance in each visible frame
• meta_expressions.json: JSON file containing referring expressions and metadata

Annotation Format

The meta_expressions.json file contains:

{
  "videos": {
    "{video_id}": {
      "frames": ["00000", "00001", ...],
      "expressions": {
        "{expression_id}": {
          "exp": "referring expression text",
          "obj_id": object_id,
          "exp_type": "static|dynamic|hybrid"
        }
      }
    }
  }
}

Usage (Please refer to the GitHub repository)

Downloading the Dataset

Option 1: Using the Download Script

python scripts/download_dataset.py \
    --repo_id iSEE-Laboratory/Long-RVOS \
    --output_dir data

Option 2: Using Hugging Face Hub API

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="iSEE-Laboratory/Long-RVOS",
    repo_type="dataset",
    local_dir="./data"
)

Option 3: Manual Download

Download from this repo or Google Drive.

Citation

If you use the Long-RVOS dataset in your research, please cite:

@article{liang2025longrvos,
  title={Long-RVOS: A Comprehensive Benchmark for Long-term Referring Video Object Segmentation},
  author={Liang, Tianming and Jiang, Haichao and Yang, Yuting and Tan, Chaolei and Li, Shuai and Zheng, Wei-Shi and Hu, Jian-Fang},
  journal={arXiv preprint arXiv:2505.12702},
  year={2025}
}

License

This dataset is licensed under the Apache 2.0 License. Please refer to the LICENSE file for details.

Contact

For questions, issues, or contributions, please refer to the GitHub repository.

---

Dataset Version: 1.0
Last Updated: 2025