A Downsampled Variant of ImageNet as an Alternative to the CIFAR datasets

By Patryk Chrabaszcz, Ilya Loshchilov, Frank Hutter. 2017

772 citations recorded by the source corpus.

The original ImageNet dataset is a popular large-scale benchmark for training Deep Neural Networks. Since the cost of performing experiments (e.g, algorithm design, architecture search, and hyperparameter tuning) on the original dataset might be prohibitive, we propose to consider a downsampled version of ImageNet. In contrast to the CIFAR datasets and earlier downsampled versions of ImageNet, our proposed ImageNet32$\times$32 (and its variants ImageNet64$\times$64 and ImageNet16$\times$16) contains exactly the same number of classes and images as ImageNet, with the only difference that the images are downsampled to 32$\times$32 pixels per image (64$\times$64 and 16$\times$16 pixels for the variants, respectively). Experiments on these downsampled variants are dramatically faster than on the original ImageNet and the characteristics of the downsampled datasets with respect to optimal hyperparameters appear to remain similar. The proposed datasets and scripts to reproduce our results are available at this http URL and this https URL

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Evaluation & Benchmarks

Citations in the corpus

Scaling Laws for Autoregressive Generative Modeling (2020, methodology): “We will show that ImageNet [CLH17] classiﬁcation performance improves smoothly with pre-trained model size, following another power law.”
Fixing Weight Decay Regularization in Adam (2017, methodology): “(7), to double-check that this is not a coincidence, we also performed experiments on the ImageNet32x32 dataset (Chrabaszcz et al., 2017), a downsampled version of the original ImageNet dataset with 1.2 million 32 × 32 pixels images, where an epoch is 24 times longer than on CIFAR-10.”
AutoML: A Survey of the State-of-the-Art (2019, background): “In [116], ImageNet64×64 and its variants 32×32, 16×16 are provided, at the same time these lower resolution dataset can”
Efficient Multi-Objective Neural Architecture Search via Lamarckian Evolution (2018, methodology): “To study the transferability of the discovered cells to a different dataset (without having to run architecture search itself on the target dataset), we built architectures suited for ImageNet64x64 (Chrabaszcz et al., 2017) based on five cells discovered on CIFAR-10.”
AutoML-Zero: Evolving Machine Learning Algorithms From Scratch (2020, methodology): “To make sure the improvement is not specific to CIFAR-10, we further show the gain generalizes to other datasets: SVHN (Netzer et al., 2011), ImageNet (Chrabaszcz et al., 2017), and Fashion MNIST (Xiao et al., 2017).”
NAS-Bench-201: Extending the Scope of Reproducible Neural Architecture Search (2020, methodology): “In sum, ImageNet-16-120 contains 151.7K training images, 3K validation images, and 3K test images with 120 classes.”
AutoGAN: Neural Architecture Search for Generative Adversarial Networks (2019, methodology): “Examples include early stop [7, 72], using low-resolution images [57, 10, 29], performance prediction with a surrogate model [37], employing a small backbone [7] or leveraging shared parameters [46].”
Using Pre-Training Can Improve Model Robustness and Uncertainty (2019, methodology): “For ablation experiments, we remove 153 CIFAR-10-related classes from the Downsampled ImageNet dataset.”
DLCFT: Deep Linear Continual Fine-Tuning for General Incremental Learning (2022, methodology): “For the experiments using CIFAR-100 dataset [19], we use ImageNet32 [6] which consists of 32×32 downsampled images of ImageNet-1k dataset [6, 39].”
Overcoming Catastrophic Forgetting With Unlabeled Data in the Wild (2019, methodology): “We evaluate the compared methods on CIFAR-100 [18] and ImageNet ILSVRC 2012 [6], where all images are downsampled to 32 × 32 [5].”
Class-incremental Learning via Deep Model Consolidation (2019, methodology): “We use ImageNet32×32 dataset [8] as the source of auxiliary data for model consolidation stage.”
Flow Matching for Generative Modeling (2022, positive): “For images, we center crop images and resize to the appropriate dimension, whereas for the 32 × 32 and 64 × 64 resolutions we use the same pre-processing as (Chrabaszcz et al., 2017).”
Deep Anomaly Detection with Outlier Exposure (2018, methodology): “The dataset’s images were cropped using bounding box information so that cropped images contain the target, unlike Downsampled ImageNet (Chrabaszcz et al., 2017).”
A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks (2018, methodology): “For the task of class-incremental learning, we train ResNet with 34 layers for classifying CIFAR-100 and downsampled ImageNet [4].”
Defining Benchmarks for Continual Few-Shot Learning (2020, methodology): “For this reason we propose a new variant of ImageNet64×64 (Chrabaszcz et al., 2017), named SlimageNet64 (derived from Slim and ImageNet).”
LeViT: a Vision Transformer in ConvNet’s Clothing for Faster Inference (2021, background): “Due to the quadratic computational complexity and number of parameters involved in attention mechanisms, most authors [23, 24] initially considered images of small sizes like in CIFAR or Imagenet64 [25].”
Automated Machine Learning: From Principles to Practices (2018, methodology): “The prediction can be achieved by traditional curve fitting methods [34] or training a surrogate model for prediction [7, 73].”
Red Alarm for Pre-trained Models: Universal Vulnerabilities by Neuron-Level Backdoor Attacks (2021, background): “(Chrabaszcz et al., 2017) and train the PTMs for 30 epochs.”