VIL-100

VIL-100数据集由天津大学创建，包含100个视频，总计10,000帧，覆盖多种实际交通场景。每帧视频均手动标注高质量实例级车道标注，支持帧级和视频级性能评估。数据集创建过程中，考虑了10种典型场景，如正常、拥挤、曲线道路等，增加了数据集的复杂性和真实性。VIL-100数据集主要应用于自动驾驶领域，旨在解决视频级车道检测问题，支持车道变更、轨迹规划和自主导航等关键功能。

The VIL-100 dataset, developed by Tianjin University, comprises 100 videos totaling 10,000 frames covering diverse real-world traffic scenarios. Each frame is manually annotated with high-quality instance-level lane annotations, supporting both frame-level and video-level performance evaluations. During its construction, 10 typical scenarios were considered, including normal driving, congested traffic, curved roads and others, to enhance the dataset's complexity and realism. Primarily applied in the field of autonomous driving, the dataset aims to address video-level lane detection tasks, and supports key functionalities such as lane changing, trajectory planning and autonomous navigation.

搜集汇总

背景与挑战

背景概述

VIL-100是一个专为视频实例车道检测设计的数据集，与ICCV 2021会议相关。该数据集提供了详细的下载、训练、测试和评估指南，支持基于MMA-Net模型的开发和研究。