Robusto-1 Dataset
收藏资源简介:
Robusto-1数据集是由Artificio和UTEC大学合作创建的,包含来自秘鲁的驾驶场景视频。该数据集选用了285段视频,并采样出200个5秒钟的视频片段用于构建数据集,另外7段视频用于初步的VQA分析。这些视频展示了具有挑战性的驾驶环境,如激烈的驾驶行为、高交通指数和大量罕见街道物体。数据集通过提出15个问题来评估人类与视觉语言模型在视觉问题回答任务上的认知对齐程度,包括变量问题、选择题和反事实假设问题。该数据集旨在推动自动驾驶系统在非常规场景下的性能评估,特别是在面临预期之外情况时的表现。
The Robusto-1 dataset was co-created by Artificio and UTEC University, containing driving scenario videos from Peru. It selects 285 video segments, from which 200 5-second video clips are sampled to construct the dataset, while another 7 videos are used for preliminary VQA analysis. These videos showcase challenging driving environments, such as aggressive driving behaviors, high traffic density, and a large number of rare street objects. The dataset proposes 15 questions to evaluate the cognitive alignment between humans and vision-language models on the visual question answering (VQA) task, including variable questions, multiple-choice questions, and counterfactual hypothetical questions. This dataset aims to promote the performance evaluation of autonomous driving systems in unconventional scenarios, especially their performance when facing unexpected situations.

- 1Robusto-1 Dataset: Comparing Humans and VLMs on real out-of-distribution Autonomous Driving VQA from PeruArtificio, Universidad de Ingeneria y Tecnologia (UTEC) · 2025年



