BigDetection

BigDetection是一个新的、大规模的基准数据集，旨在通过整合现有数据集（LVIS、OpenImages和Object365）的训练数据，并设计精细的原则，为改进目标检测器的预训练提供一个更大的数据集。该数据集包含600个对象类别，超过340万训练图像和3600万个边界框，规模远超以往的基准，为评估不同的目标检测方法和作为预训练数据集提供了新的机会和挑战。BigDetection is a novel, large-scale benchmark dataset developed to provide a larger-scale resource for enhancing the pre-training of object detectors, by integrating training data from existing datasets including LVIS, OpenImages, and Object365 alongside carefully curated principles. It consists of 600 object categories, more than 3.4 million training images and 36 million bounding boxes. With a scale significantly exceeding prior benchmarks, it presents new opportunities and challenges for evaluating diverse object detection methods and serving as a pre-training dataset.

搜集汇总

背景与挑战

背景概述

BigDetection是一个大规模基准数据集，整合了LVIS、OpenImages和Object365的训练数据，包含600个对象类别、超过340万张训练图像和3600万个边界框，规模远超以往基准，旨在为目标检测器的预训练和评估提供新的机会和挑战。