RailSem19

该数据集是一个针对轨道路径提取的基准数据集，经过调整后，仅包含可能的自主体路径三元组以供评估。此外，该数据集被修改，以便基于真实阳性轨道像素级别、所有轨道像素级别以及轨道路径级别的指标来评估路径提取算法。该数据集的规模包括6000张训练图像和1000张验证图像，其任务涉及路径提议生成和轨道路径提取。This dataset is a benchmark dataset for track path extraction. It has been adjusted to only include feasible ego-agent path triplets for evaluation purposes. Additionally, it has been modified to support the evaluation of path extraction algorithms using metrics across three levels: pixel-level metrics for true positive track pixels, pixel-level metrics for all track pixels, and track path-level metrics. The dataset comprises 6000 training images and 1000 validation images, with its targeted tasks covering path proposal generation and track path extraction.

搜集汇总

背景与挑战

背景概述

RailSem19是一个包含8500张铁路车辆视角图像的语义理解数据集，提供丰富的语义标注，适用于铁路和道路场景分析。该数据集免费公开，但需遵守相关使用许可。