IVUlab/pixcvbench

--- configs: - config_name: default data_files: - split: test path: - "Objects.csv" - "ADE20K/FinalSegmentations.json" - "COCO/FinalSegmentations.json" --- # PixCV-Bench Benchmark The dataset annotations augmenting CV-Bench with referring expressions and corresponding segmentation masks for the objects of interest in their respective questions within the original VQA task. ## Note Objects annotated with red box in the dataset are filtered using the original prompt and is not available in the referring expression csv annotations file. # Acknowledgements I acknowledge the use of CV-Bench dataset's images and questions/choices part of building this dataset, the original [CV-Bench](https://huggingface.co/datasets/nyu-visionx/CV-Bench). # References Please city my work if you find the dataset useful ``` @article{siam2025pixfoundation, title={PixFoundation: Are We Heading in the Right Direction with Pixel-level Vision Foundation Models?}, author={Siam, Mennatullah}, journal={arXiv preprint arXiv:2502.04192}, year={2025} } ```--- configs: - config_name: 默认 data_files: - split: 测试集 path: - "Objects.csv" - "ADE20K/FinalSegmentations.json" - "COCO/FinalSegmentations.json" --- # PixCV-Bench 基准测试集 本数据集的标注为CV-Bench新增了指代表达式（referring expressions），以及原始视觉问答（VQA, Visual Question Answering）任务中各问题对应关注对象的配套分割掩码（segmentation masks）。 ## 说明 数据集中以红框标注的对象已通过原始提示词完成过滤，未收录于指代表达式CSV格式的标注文件中。 # 致谢 本数据集的构建过程中使用了原始[CV-Bench](https://huggingface.co/datasets/nyu-visionx/CV-Bench)的图像、问题与选项内容。 # 参考文献 若您认为本数据集对您的研究有所帮助，请引用本文工作： @article{siam2025pixfoundation, title={PixFoundation: Are We Heading in the Right Direction with Pixel-level Vision Foundation Models?}, author={Siam, Mennatullah}, journal={arXiv preprint arXiv:2502.04192}, year={2025} }