FORB

FORB是Snap Inc.推出的一个用于平面物体检索的基准数据集，专注于多样化的2D平面物体。该数据集包含54,000张图像，旨在评估图像嵌入在分布外领域的一般化能力。FORB补充了现有的图像检索基准，并作为一个测试平台，用于评估在分布外领域图像嵌入质量。数据集涵盖了日常生活中常见的物体，如绘画、标志等，适用于多种实际视觉搜索应用，如品牌推广中的标志识别、博物馆艺术品展览的增强以及在线购物等。

FORB is a planar object retrieval benchmark dataset launched by Snap Inc., focusing on diverse 2D planar objects. This dataset contains 54,000 images, aiming to evaluate the generalization capability of image embeddings in out-of-distribution (OOD) domains. FORB complements existing image retrieval benchmarks and serves as a testbed for assessing the quality of image embeddings in out-of-distribution scenarios. The dataset covers common daily objects such as paintings, logos and the like, and is applicable to a variety of practical visual search applications, including logo recognition in brand promotion, enhancement of museum art exhibitions, and online shopping.