UrbanCars

该数据集是一个由多种来源合成的汽车数据集，包含了如背景和共同出现的物体等具有偏见的属性。图片是从斯坦福汽车、地点和LVIS数据集中分割并重新组合而成的，特别关注了那些具有偏见的属性。该数据集的规模为合成数据，包括8000张不平衡的训练图片和1000张平衡的测试图片，其任务旨在进行汽车识别和去偏见处理。

This is a synthetic automotive dataset constructed from multiple sources, incorporating biased attributes such as background contexts and co-occurring objects. All image samples are segmented and recombined from the Stanford Cars, Place, and LVIS datasets, with special focus on these biased attributes. This synthetic dataset comprises 8,000 imbalanced training images and 1,000 balanced test images, and its intended tasks are vehicle recognition and debiasing.