Syn2Real 合成图像数据集

Syn2Real 是一个合成—真实视觉领域的数据集。该数据集旨在鼓励进一步开发强大的域转移方法。其目标是训练合成 “源” 域上的模型，然后对其进行更新，以便在真正的 “目标” 域上提高其性能，而无需使用任何目标标注。数据集包括三项任务：具有已知类别集的更传统的封闭集分类任务；研究较少的开放集分类任务（目标域中对象类别未知）以及对象检测任务（涉及通过预测对象的边界框和相应的类标签来定位对象实例）。

Syn2Real is a synthetic-to-real visual domain dataset. This dataset aims to encourage further development of robust domain shift methods. Its objective is to train models on the synthetic "source" domain, then fine-tune them to improve their performance on the real "target" domain without using any target domain annotations. The dataset includes three tasks: the more traditional closed-set classification task with a known category set; the under-researched open-set classification task (where object categories in the target domain are unknown); and the object detection task, which involves localizing object instances by predicting their bounding boxes and corresponding class labels.

搜集汇总

背景与挑战

背景概述

Syn2Real是一个合成—真实视觉领域的数据集，旨在促进域转移方法的研究，包含封闭集分类、开放集分类和对象检测三项任务。该数据集由社区用户贡献，主要用于教育和信息目的。