Unconstrained Vehicle Identification Benchmark (UVIB)
收藏资源简介:
UVIB(Unconstrained Vehicle Identification Benchmark)是一个用于评估跨域监控场景下车辆属性分类的基准数据集,由巴西联邦巴拉那大学和巴拉那天主教大学联合创建。该数据集整合了七个巴西公开数据集的84,835张车辆图像,划分为监控域(57,798张)和通用域(27,037张),并提供了朝向、车辆品牌型号识别适用性及颜色清晰度三项二进制标注,这些标注在原数据集中并未统一提供。数据集的构建过程包括从各源数据集中提取车辆裁剪图像(部分借助YOLO自动检测),随后由多名标注员进行人工标注,并经过多轮审核与一致性校验以确保标注质量。该基准旨在解决模型在真实监控场景中因视角、遮挡、光照和传感器差异导致的性能退化问题,为智能交通系统中的车辆识别、交通监控和法医调查等应用提供鲁棒性评估标准。
UVIB (Unconstrained Vehicle Identification Benchmark) is a benchmark dataset for evaluating vehicle attribute classification in cross-domain surveillance scenarios, co-developed by the Federal University of Paraná and the Pontifical Catholic University of Paraná in Brazil. This dataset integrates 84,835 vehicle images from seven publicly available Brazilian datasets, which are split into two domains: the surveillance domain (57,798 images) and the general domain (27,037 images). It provides three types of binary annotations including orientation, applicability for vehicle brand and model recognition, and color clarity, which were not uniformly supplied in the original source datasets. The dataset construction process involves extracting cropped vehicle images from each source dataset (some via automatic detection using YOLO), followed by manual annotation by multiple annotators, as well as multi-round review and consistency verification to ensure annotation quality. This benchmark aims to address the performance degradation of models in real-world surveillance scenarios caused by variations in viewpoint, occlusion, illumination and sensor settings, and provide robust evaluation standards for applications such as vehicle recognition, traffic surveillance and forensic investigation in intelligent transportation systems.

- 1A Benchmark for Vehicle Attribute Classification in Cross-Domain Surveillance Scenarios联邦巴拉那大学; 巴拉那天主教大学 · 2026年




