COCO-ReM
收藏资源简介:
COCO-ReM是由印度理工学院罗基分校的研究团队开发的一个数据集,是对原有的COCO数据集的改进版本。该数据集包含1093027条高质量的实例标注,旨在提供更精确的边界和更全面的实例覆盖,以支持对象检测和分割任务。COCO-ReM通过引入更精细的标注流程,包括使用Segment Anything Model (SAM) 来改进边界质量,以及从LVIS数据集和LVIS训练的模型中导入实例来增强数据集的完整性。该数据集适用于评估和训练对象检测模型,特别是在需要高精度和快速收敛的场景中。COCO-ReM的访问地址为https://cocorem.xyz,旨在推动对象检测领域的进一步研究和开发。
COCO-ReM is an improved variant of the original COCO dataset, developed by a research team from the Indian Institute of Technology Roorkee (IIT Roorkee). It contains 1,093,027 high-quality instance annotations, aiming to provide more precise boundaries and more comprehensive instance coverage to support object detection and segmentation tasks. COCO-ReM incorporates a more meticulous annotation pipeline, including leveraging the Segment Anything Model (SAM) to enhance boundary quality, as well as importing instances from the LVIS dataset and models trained on LVIS to boost the dataset’s comprehensiveness. This dataset is applicable for evaluating and training object detection models, especially in scenarios requiring high precision and fast convergence. The official access link of COCO-ReM is https://cocorem.xyz, which is intended to advance further research and development in the field of object detection.

- 1Benchmarking Object Detectors with COCO: A New Path Forward印度理工学院罗基分校 · 2024年



