AddNIST Dataset

Dataset containing the images and labels for the AddNIST data used in the CVPR NAS workshop Unseen-data challenge under the codename "Adaline"The AddNIST dataset is a constructed dataset from MNIST Images. The intention of this dataset is to require machine learning models to do more than just image classification but also perform a calculation, in this case addition. For each image, three MNIST Images were randomly chosen and combined together through the colour channels, resulting in a three colour-channel image so each MNIST image represents one colour channel. The data is in a channels-first format with a shape of (n, 3, 28, 28) where n is the number of samples in the corresponding set (45,000 for training, 15,000 for validation, and 10,000 for testing).There are twenty classes in the dataset, with 3,500 examples of each, distributed evenly between the three subsets.The label of each image is generated using the formula "(r + b + g) - 1" where r, g, and b are the red, green, and blue colour channels respectively. An example of an AddNIST Image would be a rgb configuation of 3, 7, and 4 respectively, which would result in a label of 13 ((3 + 7 + 4) - 1).

本数据集包含代号为"Adaline"的CVPR神经架构搜索（Neural Architecture Search）研讨会「不可见数据挑战赛」所使用的AddNIST数据集的图像与标签。AddNIST数据集是由MNIST图像构建而成的人工构建数据集。该数据集的设计初衷并非仅要求机器学习模型完成图像分类任务，而是需其同时执行计算操作——本任务为加法运算。对于每张图像，我们随机选取三张MNIST图像，通过颜色通道将其组合为一张三通道彩色图像，每张MNIST图像分别对应红、绿、蓝其中一个颜色通道。数据采用通道优先（channels-first）格式，形状为(n, 3, 28, 28)，其中n为对应子集的样本量：训练集包含45000个样本，验证集15000个，测试集10000个。该数据集共设20个类别，每类含3500个样本，且在训练、验证、测试三个子集上均匀分布。每张图像的标签通过公式"(r + b + g) - 1"生成，其中r、g、b分别代表红色、绿色、蓝色通道的数值。例如，若某图像的RGB通道数值依次为3、7、4，则其标签为13，即(3 + 7 + 4) - 1。