IDDA

IDDA是由都灵理工大学的研究团队创建的一个大规模合成数据集，专为自动驾驶中的语义分割任务设计。该数据集包含超过100万个标注图像，涵盖超过100种不同的视觉域，特别关注于解决训练和测试数据之间的域偏移问题。IDDA通过模拟多种天气和视角条件，以及七个不同城市类型，支持对当前和未来最先进的语义分割架构进行深入分析和基准测试。数据集的创建过程利用了CARLA模拟器，确保了数据的高质量和多样性，适用于单源或多源域适应技术的研究和评估。IDDA is a large-scale synthetic dataset created by a research team from Politecnico di Torino, specifically designed for semantic segmentation tasks in autonomous driving. This dataset contains over 1 million annotated images, covering more than 100 distinct visual domains, with a particular focus on mitigating the domain shift problem between training and test data. IDDA supports in-depth analysis and benchmarking of current and state-of-the-art semantic segmentation architectures by simulating diverse weather and viewpoint conditions, as well as seven different urban types. The dataset was developed using the CARLA simulator, ensuring high data quality and diversity, and is suitable for research and evaluation of single-source or multi-source domain adaptation techniques.

搜集汇总

背景与挑战

背景概述

IDDA是一个大型合成数据集，专为语义分割任务设计，包含超过一百万张Full HD图像，覆盖105个不同场景，可用于评估模型的泛化能力和域适应性能。