BENCHMARK.md

ReasonVOS Benchmark

To effectively evaluate the reasoning segmentation task in videos, we construct a new benchmark, called ReasonVOS. Note that this dataset is constructed based on existing video object segmentation datasets, including BURST (released under MIT License), VIPSeg (released under non-commercial research purpose only), MOSE (released under CC BY-NC-SA 4.0 License), and MeViS (released under CC BY-NC-SA 4.0 License).

Shout out to these amazing works!!

License Agreement

⚠️⚠️⚠️ By downloading, using, or sharing this dataset, you agree to comply with the licenses outlined above.

Download

Google Drive: https://drive.google.com/file/d/1QMggGR91AE-iUUub5YwzMYiFt0TeV5_8/view?usp=sharing

Please refer to benchmark_check.py to explore the structure of the dataset.

Statistics

License

The data of ReasonVOS is released for non-commercial research purpose only.