Syn2Real

Syn2Real数据集是由波士顿大学创建的一个大规模视觉领域适应性基准数据集，旨在促进从合成图像到真实图像的模型转移研究。该数据集包含超过280,000张图像，涵盖12个对象类别，分为合成源域和两个真实目标域。数据集设计了三个相关任务：封闭集对象分类、开放集对象分类和对象检测。通过这个数据集，研究者可以评估和开发在合成到真实图像转换中表现优异的适应方法，解决在机器人、医学成像和监控等领域中遇到的挑战。The Syn2Real Dataset is a large-scale visual domain adaptation benchmark dataset created by Boston University, which aims to facilitate research on model transfer from synthetic images to real-world images. It contains over 280,000 images spanning 12 object categories, and is partitioned into a synthetic source domain and two real target domains. Three related tasks are designed for this dataset: closed-set object classification, open-set object classification, and object detection. With this dataset, researchers can evaluate and develop adaptation methods that perform excellently in synthetic-to-real image translation scenarios, addressing challenges encountered in fields such as robotics, medical imaging, and surveillance.

搜集汇总

背景与挑战

背景概述

Syn2Real是一个大规模合成到真实视觉域适应基准数据集，包含超过280K图像，涵盖闭集分类、开集分类和物体检测三个任务。它采用无监督域适应方法，使用合成源域和两个真实目标域，旨在促进跨域模型适应方法的发展，特别是在开集和检测任务上仍存在较大挑战。