EMNIST: an extension of MNIST to handwritten letters

By Gregory Cohen, Saeed Afshar, Jonathan C. Tapson, A. van Schaik. 2017

809 citations recorded by the source corpus.

The MNIST dataset has become a standard benchmark for learning, classification and computer vision systems. Contributing to its widespread adoption are the understandable and intuitive nature of the task, its relatively small size and storage requirements and the accessibility and ease-of-use of the database itself. The MNIST database was derived from a larger dataset known as the NIST Special Database 19 which contains digits, uppercase and lowercase handwritten letters. This paper introduces a variant of the full NIST dataset, which we have called Extended MNIST (EMNIST), which follows the same conversion paradigm used to create the MNIST dataset. The result is a set of datasets that constitute a more challenging classification tasks involving letters and digits, and that shares the same image structure and parameters as the original MNIST task, allowing for direct compatibility with all existing classifiers and systems. Benchmark results are presented along with a validation of the conversion process through the comparison of the classification results on converted NIST digits and the MNIST digits.

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

Federated Optimization in Heterogeneous Networks (2018, methodology): “• FEMNIST: We study an image classification problem on the 62-class EMNIST dataset (Cohen et al., 2017) using multinomial logistic regression.”
FedML: A Research Library and Benchmark for Federated Machine Learning (2020, methodology): “The federated version of EMNIST [72] partitions the digits by their author.”
Robust Aggregation for Federated Learning (2019, methodology): “1 Handwritten-Letter Recognition The first dataset is the EMNIST dataset [26] for handwritten letter recognition.”
Measuring the Effects of Non-Identical Data Distribution for Federated Visual Classification (2019, methodology): “For example, Caldas et al. [2018] use Extended MNIST [Cohen et al., 2017] with partitions over writers of the digits, rather than simply partitioning over digit class.”
Expanding the Reach of Federated Learning by Reducing Client Resource Requirements (2018, methodology): “Datasets We use three datasets in our experiments: MNIST [LeCun et al., 1998], CIFAR-10 [Krizhevsky and Hinton, 2009] and Extended MNIST or EMNIST [Cohen et al., 2017].”
Knowledge Flow: Improve Upon Your Teachers (2019, methodology): “For supervised learning, we use a variety of image classiﬁcation benchmarks, including CIFAR-10 (Krizhevsky, 2009), CIFAR-100 (Krizhevsky, 2009), STL-10 (Coates et al., 2011), and EM-NIST (Cohen et al., 2017).”
Adaptive Orthogonal Projection for Batch and Online Continual Learning (2022, methodology): “It gets only 25.42 on EMNIST-47, much worse than AOP (80.97) with 24 tasks.”
NISPA: Neuro-Inspired Stability-Plasticity Adaptation for Continual Learning in Sparse Networks (2022, methodology): “In all datasets (CIFAR10, CIFAR100, MNIST, EMNIST, and FashionMNIST), we report the accuracy on the ofﬁcial test dataset and use 10% of the training dataset for validation.”
Continual Learning by Using Information of Each Class Holistically (2021, methodology): “Comparing the performance in this setting with the 1 class per task setting of PCL, we see that PCL works better in Five classes per task : We use EMNIST-47 and CIFAR100 to test 5 classes per task as they have a large number of classes to form many tasks.”
Overcoming Catastrophic Forgetting by Neuron-level Plasticity Control (2019, methodology): “1998), fashion-MNIST(fMNIST)(Xiao, Rasul, and Vollgraf 2017), EMNIST(balanced dataset)(Cohen et al. 2017), and smallNORB (LeCun 2004).”
Why ReLU Networks Yield High-Confidence Predictions Far Away From the Training Data and How to Mitigate the Problem (2018, methodology): “…the conﬁdence on the out-distribution, we use two datasets on CIFAR10 [17] and SVHN [27] (namely among CIFAR10, SVHN and the classroom subset of LSUN [36] we use the one on which we have not trained) and for MNIST we evaluate on EMNIST [7], a grayscale version of CIFAR10 and Fashion MNIST [35].”
Mixout: Effective Regularization to Finetune Large-scale Pretrained Language Models (2019, methodology): “We monitor (cid:107) w ft − w pre (cid:107) 2 , 1 validation accuracy on MNIST, and validation accuracy on EMNIST Digits to compare mixout ( w pre , p ) to dropout( p ) across 10 random restarts.”
Disentanglement by Nonlinear ICA with General Incompressible-flow Networks (GIN) (2020, methodology): “Experiments on the EMNIST dataset (Cohen et al., 2017) demonstrate the ability of GIN to estimate independent and interpretable latent variables from real-world data.”
Unsupervised Progressive Learning and the STAM Architecture (2019, neutral): “We include results on four datasets: MNIST [Lecun et al. , 1998] , EMNIST (balanced split with 47 classes) [Cohen et al. , 2017] , SVHN [Netzer et al. , 2011] , and CIFAR-10 [Krizhevsky et al. , 2014] .”
Disentangling Adversarial Robustness and Generalization (2018, methodology): “, class manifolds “5” and “6” on EMNIST [19], allow to study their relation to generalization.”