Mapillary Vistas Dataset

The Mapillary Vistas Dataset is a substantial, street-level image dataset containing 25,000 high-resolution images annotated across 124 classes (70 instance-specific, 46 stuff, 8 void or crowd). Annotation adopts a dense, fine-grained style using polygons to delineate individual objects. The dataset, authored by the dataset creators, is notably larger, by a factor of 5, than the combined fine annotations in Cityscapes. It encompasses images captured worldwide, encompassing diverse weather, seasonal, and daytime conditions. The images are sourced from various devices such as mobile phones, tablets, action cameras, professional capturing rigs, and different experienced photographers, thus embracing diversity, detail richness, and global coverage.Mapillary Vistas 数据集系一项规模庞大的街道级图像数据集，包含标注于 124 个类别（其中 70 个为实例特定，46 个为物体类别，8 个为空或人群类别）的 25,000 张高分辨率图像。标注采用密集且细致的风格，通过多边形勾勒出单个物体。该数据集由数据集制作者所创建，其规模显著超过 Cityscapes 中所有精细标注的总和，达到其 5 倍之大。该数据集涵盖了全球范围内的图像，包括不同的天气、季节和白天状况。图像来源于多种设备，如移动电话、平板电脑、运动相机、专业拍摄设备以及不同经验水平的摄影师，因此融合了多样性、细节丰富性和全球覆盖范围。

搜集汇总

背景与挑战

背景概述

Mapillary Vistas Dataset是一个大规模街景图像数据集，包含25,000张高分辨率图像，覆盖全球多样环境条件，适用于自动驾驶中的语义分割和实例分割任务。数据集具有124个细粒度类别，采用多边形标注，提供训练、验证和测试分割，并遵循CC BY-NC-SA 4.0许可证。