InLeaf Dataset

We have created Inleaf dataset for which the images are downloaded from the internet. It is less expensive to create such a dataset as opposed to well-defined standard datasets, which are often collected by capturing images from a high-quality camera and/or processed in the lab before the training. In addition, the dataset generated from the internet reflects realistic images, which includes plenty of mislabeled, noisy and irrelevant images.我们构建了Inleaf数据集，其图像均源自互联网下载。相较于需借助高质量相机拍摄图像，并在训练前于实验室完成预处理的标准化基准数据集，此类数据集的构建成本显著更低。此外，该互联网采集得到的数据集涵盖真实场景图像，其中包含大量标注错误、带噪声以及不相关的图像样本。