COCO_OI, ObjectNet_D

COCO_OI和ObjectNet_D是针对COCO数据集的补充，旨在解决现有数据集在训练深度模型时可能出现的饱和问题。COCO_OI由COCO和OpenImages的共同类别图像组成，包含1,418,978个训练边界框和41,893个验证边界框，用于测试对象检测模型的泛化能力。ObjectNet_D则专注于日常生活中的物体，包含5875个边界框，用于评估模型在不同视角和背景下的表现。这两个数据集的创建旨在提供更多样化的数据，以增强模型的鲁棒性和泛化能力。COCO_OI and ObjectNet_D are supplementary datasets to the COCO dataset, designed to address the potential saturation issue that may arise when training deep models with existing datasets. COCO_OI comprises images from shared categories of COCO and OpenImages, containing 1,418,978 training bounding boxes and 41,893 validation bounding boxes, serving to test the generalization ability of object detection models. ObjectNet_D focuses on everyday objects, including 5,875 bounding boxes, and is used to evaluate model performance under different viewpoints and backgrounds. The development of these two datasets aims to provide more diverse data to enhance the robustness and generalization ability of models.