Objects365 v1

Objects365 Dataset V1 是一个全新的数据集，旨在促进目标检测研究，重点关注野外的各种物体。 它有超过 600K 训练图像的 365 个对象类别。 超过 1000 万个高质量的边界框是通过精心设计的三步注释管道手动标记的。 它是迄今为止最大的对象检测数据集（具有完整注释），并为社区建立了更具挑战性的基准。 Objects365 可以作为更好的特征学习数据集，用于对象检测和语义分割等本地化敏感任务。
Objects365 Consortium 不拥有图像的版权。 图片的使用必须遵守 Flickr 使用条款。 图像的用户对数据集的使用承担全部责任，包括但不限于使用他们可能从数据集创建的受版权保护的图像的任何副本。 您不会分发上述图片。 如果侵犯图片版权，我们将立即删除图片。

Objects365 Dataset V1 is a novel dataset developed to advance object detection research, focusing on diverse objects in the wild. It encompasses 365 object categories across over 600,000 training images. More than 10 million high-quality bounding boxes were manually annotated via a meticulously designed three-step annotation pipeline. It is the largest fully annotated object detection dataset to date, and has established a more challenging benchmark for the research community. Objects365 can serve as an excellent feature learning dataset for localization-sensitive tasks such as object detection and semantic segmentation.

The Objects365 Consortium does not hold the copyright of the included images. The use of these images must comply with the Flickr Terms of Use. Users of the dataset bear full responsibility for their usage of the dataset, including but not limited to any copyrighted image copies they may create from the dataset. You are prohibited from redistributing the aforementioned images. We will immediately remove the images upon detection of any copyright infringement.