BDD-X (Berkeley Deep Drive-X (eXplanation))

Berkeley Deep Drive-X (eXplanation) 是一个数据集，由 6,970 个视频中超过 77 小时的驾驶组成。这些视频是在不同的驾驶条件下拍摄的，例如白天/夜晚、高速公路/城市/乡村、夏季/冬季等。平均时长 40 秒，每个视频包含大约 3-4 个动作，例如加速、减速、右转等.，所有这些都带有描述和解释的注释。我们的数据集包含超过 840 万帧中的超过 26K 活动。

Berkeley Deep Drive-X (eXplanation) is a dataset consisting of over 77 hours of driving footage from 6,970 video clips. These videos were captured under diverse driving conditions, including daytime/nighttime, highway/urban/rural environments, summer/winter weather, and more. Each video has an average duration of 40 seconds and contains approximately 3-4 driving actions such as acceleration, deceleration, right turns, and others. All these actions are annotated with descriptive and explanatory notes. Our dataset includes over 26,000 activities across more than 8.4 million frames.

搜集汇总

背景与挑战

背景概述

BDD-X是一个自动驾驶领域的数据集，包含6,970个视频（总计超过77小时），覆盖白天/夜晚、高速公路/城市/乡村等多种驾驶条件，每个视频平均40秒并标注了3-4个动作（如加速、减速）及其解释。该数据集旨在支持自动驾驶车辆的行为克隆和可解释人工智能研究，提供超过840万帧和26K活动的详细注释。