ObjectNet 大规模偏差控制数据集

ObjectNet 是一个通过众包形式收集的大型图像测试集。由于 ObjectNet 的图像均来自真实世界，因此目标物体形态各异，背景、旋转角度及成像视角都是随机的，这可能会严重降低识别效果。

ObjectNet is a large-scale crowdsourced image test set. As all its images are derived from real-world scenarios, the target objects exhibit varied morphologies, with their backgrounds, rotation angles and imaging perspectives fully randomized, which can severely degrade object recognition performance.