VALERIE22

VALERIE22数据集是由英特尔实验室开发的，旨在通过深度神经网络（DNNs）理解特定领域因素对感知性能的影响。该数据集通过VALERIE程序工具管道生成，提供了一个具有丰富元数据注释的逼真传感器模拟，从自动合成的场景中渲染出来。数据集包含独特的元数据集，允许提取特定的场景和语义特征，如像素级遮挡率、场景中的位置以及与相机的距离和角度。这使得可以在数据上进行多种可能的测试，并希望激发对DNNs性能理解的研究。VALERIE22数据集的应用领域主要集中在自动驾驶中的行人检测，旨在解决机器学习感知功能的安全性问题。

The VALERIE22 dataset was developed by Intel Labs to investigate the impact of domain-specific factors on perceptual performance via deep neural networks (DNNs). Generated through the VALERIE programmatic tool pipeline, this dataset provides a realistic sensor simulation richly annotated with metadata, rendered from fully automatically synthesized scenes. The dataset features a unique meta-dataset that enables extraction of targeted scene and semantic characteristics, including pixel-level occlusion rate, in-scene position, distance from the camera, and relative angle to the camera. This allows for a wide range of configurable tests on the data, and aims to inspire research focused on understanding DNN performance. The primary application domain of the VALERIE22 dataset is pedestrian detection in autonomous driving, with the goal of addressing safety-related challenges for machine learning-based perceptual functions.

搜集汇总

背景与挑战

背景概述

VALERIE22数据集由英特尔实验室开发，通过VALERIE工具管道生成逼真传感器模拟数据，包含丰富的元数据注释（如遮挡率、位置和角度），旨在研究深度神经网络在自动驾驶行人检测中的感知性能影响因素，以提升安全性。