ImageNet32x32, ImageNet16x16, ImageNet64x64
收藏资源简介:
ImageNet32x32、ImageNet16x16和ImageNet64x64是由弗莱堡大学创建的ImageNet数据集的降采样版本,分别包含32x32、16x16和64x64像素的图像。这些数据集保留了原始ImageNet的1000个类别和图像数量,仅图像分辨率降低。创建过程涉及将原始图像通过不同的降采样技术处理,并保持与CIFAR数据集相同的文件格式。这些数据集主要用于深度学习中的快速实验,如算法设计、神经网络架构搜索和超参数优化,旨在解决大规模数据集计算成本高昂的问题。
ImageNet32x32, ImageNet16x16, and ImageNet64x64 are downsampled variants of the ImageNet dataset developed by the University of Freiburg, with image resolutions of 32x32, 16x16, and 64x64 pixels respectively. These datasets preserve the 1000 categories and total number of images from the original ImageNet, with only the image resolution being reduced. The creation workflow entails processing the original images using various downsampling techniques, while retaining the identical file format as the CIFAR datasets. These datasets are primarily utilized for rapid deep learning experiments including algorithm design, neural architecture search, and hyperparameter optimization, aiming to address the high computational cost challenge posed by large-scale datasets.




