CINIC-10 is not ImageNet or CIFAR-10

TL;DR

CINIC-10 is a new dataset combining CIFAR-10 and ImageNet images, offering an extended benchmark for image classification tasks with detailed compilation and benchmarking information.

Contribution

The paper introduces CINIC-10 as an extended CIFAR-10 alternative, including compilation methodology, dataset analysis, and baseline benchmarks.

Findings

CINIC-10 dataset combines CIFAR-10 and ImageNet images.

Standard models achieve baseline performance on CINIC-10.

Dataset details and benchmarks are publicly available.

Abstract

In this brief technical report we introduce the CINIC-10 dataset as a plug-in extended alternative for CIFAR-10. It was compiled by combining CIFAR-10 with images selected and downsampled from the ImageNet database. We present the approach to compiling the dataset, illustrate the example images for different classes, give pixel distributions for each part of the repository, and give some standard benchmarks for well known models. Details for download, usage, and compilation can be found in the associated github repository.