Talk2Car

Talk2Car 数据集处于各种研究领域的交叉点，促进了跨学科解决方案的开发，以提高将自然语言融入视觉空间的最新技术。收集注释时考虑了以下方面：
自由形式的高质量自然语言命令，可刺激可在野外运行的解决方案的开发。
一个现实的任务设置。具体来说，作者考虑了一种自动驾驶环境，乘客可以通过用自然语言发出命令来控制自动驾驶汽车的动作。
Talk2Car 数据集建立在 nuScenes 数据集之上，包括一套广泛的传感器模式，即语义地图、GPS、激光雷达、雷达和带有 3D 边界框注释的 360 度 RGB 图像。如此多样的输入模式将 Talk2Car 数据集上的对象推荐任务与相关挑战区分开来，在这些挑战中，通常缺少额外的传感器模式。

The Talk2Car dataset lies at the intersection of multiple research fields, facilitating the development of interdisciplinary solutions to advance the state-of-the-art in integrating natural language into visual-spatial applications.
Annotations were collected with the following aspects in mind:
High-quality free-form natural language commands that can stimulate the development of solutions capable of operating in the wild.
A realistic task setup. Specifically, the authors considered an autonomous driving scenario where passengers can control the actions of the self-driving vehicle by issuing natural language commands.
The Talk2Car dataset is built upon the nuScenes dataset and includes a comprehensive set of sensor modalities: semantic maps, GPS, LiDAR, radar, and 360-degree RGB images annotated with 3D bounding boxes.
This diverse range of input modalities distinguishes the object referral task on the Talk2Car dataset from related challenges, which typically lack such additional sensor modalities.

搜集汇总

背景与挑战

背景概述

Talk2Car是一个专注于自动驾驶领域的数据集，旨在通过自然语言命令控制车辆动作，促进跨学科研究。它基于nuScenes数据集构建，包含多种传感器模式（如语义地图、GPS、激光雷达、雷达和360度RGB图像），以支持对象推荐任务，区别于其他仅依赖单一模式的挑战。数据集提供自由形式的高质量自然语言注释，适用于现实自动驾驶环境中的视觉-语言交互研究。