Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms

08/25/2017
∙
by   Han Xiao, et al.
∙
Zalando
∙
0
∙

We present Fashion-MNIST, a new dataset comprising of 28x28 grayscale images of 70,000 fashion products from 10 categories, with 7,000 images per category. The training set has 60,000 images and the test set has 10,000 images. Fashion-MNIST is intended to serve as a direct drop-in replacement for the original MNIST dataset for benchmarking machine learning algorithms, as it shares the same image size, data format and the structure of training and testing splits. The dataset is freely available at https://github.com/zalandoresearch/fashion-mnist

READ FULL TEXT VIEW PDF

Authors

page 3

Related Research

0
research
∙ 05/19/2022

Oracle-MNIST: a Realistic Image Dataset for Benchmarking Machine Learning Algorithms

We introduce the Oracle-MNIST dataset, comprising of 28×28 grayscale ima...
Mei Wang, et al. ∙
0
research
∙ 06/19/2019

Training on test data: Removing near duplicates in Fashion-MNIST

MNIST and Fashion MNIST are extremely popular for testing in the machine...
Christopher Geier, et al. ∙
0
research
∙ 06/22/2022

The ArtBench Dataset: Benchmarking Generative Models with Artworks

We introduce ArtBench-10, the first class-balanced, high-quality, cleanl...
Peiyuan Liao, et al. ∙
0
research
∙ 11/13/2019

IStego100K: Large-scale Image Steganalysis Dataset

In order to promote the rapid development of image steganalysis technolo...
Zhongliang Yang, et al. ∙
0
research
∙ 09/28/2020

Afro-MNIST: Synthetic generation of MNIST-style datasets for low-resource languages

We present Afro-MNIST, a set of synthetic MNIST-style datasets for four ...
Daniel J Wu, et al. ∙
0
research
∙ 05/25/2019

Cold Case: The Lost MNIST Digits

Although the popular MNIST dataset [LeCun et al., 1994] is derived from ...
Chhavi Yadav, et al. ∙
24
research
∙ 10/29/2019

Distribution Density, Tails, and Outliers in Machine Learning: Metrics and Applications

We develop techniques to quantify the degree to which a given (training ...
Nicholas Carlini, et al. ∙

Code Repositories

fashion-mnist

A MNIST-like fashion product database. Benchmark :point_down:


view repo
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1 Introduction

The MNIST dataset comprising of 10-class handwritten digits, was first introduced by LeCun et al. (1998)

in 1998. At that time one could not have foreseen the stellar rise of deep learning techniques and their performance. Despite the fact that today deep learning can do so much the simple MNIST dataset has become the most widely used testbed in deep learning, surpassing CIFAR-10 

(Krizhevsky and Hinton, 2009)

and ImageNet 

(Deng et al., 2009) in its popularity via Google trends111https://trends.google.com/trends/explore?date=all&q=mnist,CIFAR,ImageNet. Despite its simplicity its usage does not seem to be decreasing despite calls for it in the deep learning community.

The reason MNIST is so popular has to do with its size, allowing deep learning researchers to quickly check and prototype their algorithms. This is also complemented by the fact that all machine learning libraries (e.g. scikit-learn) and deep learning frameworks (e.g. Tensorflow, Pytorch) provide helper functions and convenient examples that use MNIST out of the box.

Our aim with this work is to create a good benchmark dataset which has all the accessibility of MNIST, namely its small size, straightforward encoding and permissive license. We took the approach of sticking to the classes grayscale images in the size of as in the original MNIST. In fact, the only change one needs to use this dataset is to change the URL from where the MNIST dataset is fetched. Moreover, Fashion-MNIST poses a more challenging classification task than the simple MNIST digits data, whereas the latter has been trained to accuracies above 99.7% as reported in Wan et al. (2013); Ciregan et al. (2012).

We also looked at the EMNIST dataset provided by Cohen et al. (2017)

, an extended version of MNIST that extends the number of classes by introducing uppercase and lowercase characters. However, to be able to use it seamlessly one needs to not only extend the deep learning framework’s MNIST helpers, but also change the underlying deep neural network to classify these extra classes.

2 Fashion-MNIST Dataset

Fashion-MNIST is based on the assortment on Zalando’s website222Zalando is the Europe’s largest online fashion platform. http://www.zalando.com. Every fashion product on Zalando has a set of pictures shot by professional photographers, demonstrating different aspects of the product, i.e. front and back looks, details, looks with model and in an outfit. The original picture has a light-gray background (hexadecimal color: #fdfdfd) and stored in JPEG format. For efficiently serving different frontend components, the original picture is resampled with multiple resolutions, e.g. large, medium, small, thumbnail and tiny.

We use the front look thumbnail images of unique products to build Fashion-MNIST. Those products come from different gender groups: men, women, kids and neutral. In particular, white-color products are not included in the dataset as they have low contrast to the background. The thumbnails () are then fed into the following conversion pipeline, which is visualized in Figure 1.

  1. Converting the input to a PNG image.

  2. Trimming any edges that are close to the color of the corner pixels. The “closeness” is defined by the distance within of the maximum possible intensity in RGB space.

  3. Resizing the longest edge of the image to by subsampling the pixels, i.e. some rows and columns are skipped over.

  4. Sharpening pixels using a Gaussian operator of the radius and standard deviation of

    , with increasing effect near outlines.

  5. Extending the shortest edge to and put the image to the center of the canvas.

  6. Negating the intensities of the image.

  7. Converting the image to 8-bit grayscale pixels.

Figure 1: Diagram of the conversion process used to generate Fashion-MNIST dataset. Two examples from dress and sandals categories are depicted, respectively. Each column represents a step described in section 2.
Name Description # Examples Size
train-images-idx3-ubyte.gz Training set images MBytes
train-labels-idx1-ubyte.gz Training set labels Bytes
t10k-images-idx3-ubyte.gz Test set images MBytes
t10k-labels-idx1-ubyte.gz Test set labels Bytes
Table 1: Files contained in the Fashion-MNIST dataset.

For the class labels, we use the silhouette code of the product. The silhouette code is manually labeled by the in-house fashion experts and reviewed by a separate team at Zalando. Each product contains only one silhouette code. Table 2 gives a summary of all class labels in Fashion-MNIST with examples for each class.

Finally, the dataset is divided into a training and a test set. The training set receives a randomly-selected

examples from each class. Images and labels are stored in the same file format as the MNIST data set, which is designed for storing vectors and multidimensional matrices. The result files are listed in

Table 1. We sort examples by their labels while storing, resulting in smaller label files after compression comparing to the MNIST. It is also easier to retrieve examples with a certain class label. The data shuffling job is therefore left to the algorithm developer.

Label Description Examples
T-Shirt/Top
Trouser
Pullover
Dress
Coat
Sandals
Shirt
Sneaker
Bag
Ankle boots
Table 2: Class names and example images in Fashion-MNIST dataset.

3 Experiments

We provide some classification results in Table 3 to form a benchmark on this data set. All algorithms are repeated times by shuffling the training data and the average accuracy on the test set is reported. The benchmark on the MNIST dataset is also included for a side-by-side comparison. A more comprehensive table with explanations on the algorithms can be found on https://github.com/zalandoresearch/fashion-mnist.

Test Accuracy
Classifier Parameter Fashion MNIST
DecisionTreeClassifier criterion=entropy max_depth= splitter=best
criterion=entropy max_depth= splitter=random
criterion=entropy max_depth= splitter=best
criterion=entropy max_depth= splitter=best
criterion=gini max_depth= splitter=best
criterion=entropy max_depth= splitter=random
criterion=entropy max_depth= splitter=random
criterion=gini max_depth= splitter=best
criterion=gini max_depth= splitter=best
criterion=gini max_depth= splitter=random
criterion=gini max_depth= splitter=random
criterion=gini max_depth= splitter=random
ExtraTreeClassifier criterion=gini max_depth= splitter=best
criterion=entropy max_depth= splitter=best
criterion=entropy max_depth= splitter=best
criterion=entropy max_depth= splitter=best
criterion=gini max_depth= splitter=best
criterion=gini max_depth= splitter=best
criterion=entropy max_depth= splitter=random
criterion=entropy max_depth= splitter=random
criterion=gini max_depth= splitter=random
criterion=gini max_depth= splitter=random
criterion=gini max_depth= splitter=random
criterion=entropy max_depth= splitter=random
GaussianNB priors=[0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1]
GradientBoostingClassifier n_estimators= loss=deviance max_depth=
n_estimators= loss=deviance max_depth=
n_estimators= loss=deviance max_depth=
n_estimators= loss=deviance max_depth=
n_estimators= loss=deviance max_depth=
n_estimators= loss=deviance max_depth=
n_estimators= loss=deviance max_depth=
KNeighborsClassifier weights=distance n_neighbors= p=
weights=distance n_neighbors= p=
weights=uniform n_neighbors= p=
weights=uniform n_neighbors= p=
weights=distance n_neighbors= p=
weights=distance n_neighbors= p=
weights=uniform n_neighbors= p=
weights=uniform n_neighbors= p=
weights=distance n_neighbors= p=
weights=uniform n_neighbors= p=
weights=uniform n_neighbors= p=
weights=distance n_neighbors= p=
LinearSVC loss=hinge C= multi_class=ovr penalty=l2
loss=hinge C= multi_class=crammer_singer penalty=l2
loss=squared_hinge C= multi_class=crammer_singer penalty=l2
loss=squared_hinge C= multi_class=crammer_singer penalty=l1
loss=hinge C= multi_class=crammer_singer penalty=l1
loss=squared_hinge C= multi_class=ovr penalty=l2
loss=squared_hinge C= multi_class=ovr penalty=l2
loss=squared_hinge C= multi_class=ovr penalty=l2
loss=hinge C= multi_class=ovr penalty=l2
loss=hinge C= multi_class=ovr penalty=l2
loss=hinge C= multi_class=crammer_singer penalty=l1
loss=hinge C= multi_class=crammer_singer penalty=l2
loss=squared_hinge C= multi_class=crammer_singer penalty=l2
loss=squared_hinge C= multi_class=crammer_singer penalty=l1
loss=hinge C= multi_class=crammer_singer penalty=l1
loss=hinge C= multi_class=crammer_singer penalty=l2
loss=squared_hinge C= multi_class=crammer_singer penalty=l1
loss=squared_hinge C= multi_class=crammer_singer penalty=l2
LogisticRegression C= multi_class=ovr penalty=l1
C= multi_class=ovr penalty=l2
C= multi_class=ovr penalty=l2
C= multi_class=ovr penalty=l1
C= multi_class=ovr penalty=l2
MLPClassifier activation=relu hidden_layer_sizes=[100]
activation=relu hidden_layer_sizes=[100, 10]
activation=tanh hidden_layer_sizes=[100]
activation=tanh hidden_layer_sizes=[100, 10]
activation=relu hidden_layer_sizes=[10, 10]
activation=relu hidden_layer_sizes=[10]
activation=tanh hidden_layer_sizes=[10, 10]
activation=tanh hidden_layer_sizes=[10]
PassiveAggressiveClassifier C=
C=
C=
Perceptron penalty=l1
penalty=l2
penalty=elasticnet
RandomForestClassifier n_estimators= criterion=entropy max_depth=
n_estimators= criterion=gini max_depth=
n_estimators= criterion=entropy max_depth=
n_estimators= criterion=entropy max_depth=
n_estimators= criterion=entropy max_depth=
n_estimators= criterion=gini max_depth=
n_estimators= criterion=gini max_depth=
n_estimators= criterion=gini max_depth=
n_estimators= criterion=entropy max_depth=
n_estimators= criterion=entropy max_depth=
n_estimators= criterion=gini max_depth=
n_estimators= criterion=gini max_depth=
n_estimators= criterion=entropy max_depth=
n_estimators= criterion=entropy max_depth=
n_estimators= criterion=gini max_depth=
n_estimators= criterion=gini max_depth=
n_estimators= criterion=entropy max_depth=
n_estimators= criterion=gini max_depth=
SGDClassifier loss=hinge penalty=l2
loss=perceptron penalty=l1
loss=modified_huber penalty=l1
loss=modified_huber penalty=l2
loss=log penalty=elasticnet
loss=hinge penalty=elasticnet
loss=squared_hinge penalty=elasticnet
loss=hinge penalty=l1
loss=log penalty=l1
loss=perceptron penalty=l2
loss=perceptron penalty=elasticnet
loss=squared_hinge penalty=l2
loss=modified_huber penalty=elasticnet
loss=log penalty=l2
loss=squared_hinge penalty=l1
SVC C= kernel=rbf
C= kernel=poly
C= kernel=poly
C= kernel=rbf
C= kernel=rbf
C= kernel=poly
C= kernel=linear
C= kernel=linear
C= kernel=linear
C= kernel=sigmoid
C= kernel=sigmoid
C= kernel=sigmoid
Table 3: Benchmark on Fashion-MNIST (Fashion) and MNIST.

4 Conclusions

This paper introduced Fashion-MNIST, a fashion product images dataset intended to be a drop-in replacement of MNIST and whilst providing a more challenging alternative for benchmarking machine learning algorithm. The images in Fashion-MNIST are converted to a format that matches that of the MNIST dataset, making it immediately compatible with any machine learning package capable of working with the original MNIST dataset.

References

  • Ciregan et al. [2012] D. Ciregan, U. Meier, and J. Schmidhuber. Multi-column deep neural networks for image classification. In Computer Vision and Pattern Recognition (CVPR), 2012 IEEE Conference on, pages 3642–3649. IEEE, 2012.
  • Cohen et al. [2017] G. Cohen, S. Afshar, J. Tapson, and A. van Schaik. Emnist: an extension of mnist to handwritten letters. arXiv preprint arXiv:1702.05373, 2017.
  • Deng et al. [2009] J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei. Imagenet: A large-scale hierarchical image database. In Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on, pages 248–255. IEEE, 2009.
  • Krizhevsky and Hinton [2009] A. Krizhevsky and G. Hinton. Learning multiple layers of features from tiny images. 2009.
  • LeCun et al. [1998] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998.
  • Wan et al. [2013] L. Wan, M. Zeiler, S. Zhang, Y. L. Cun, and R. Fergus. Regularization of neural networks using dropconnect. In Proceedings of the 30th international conference on machine learning (ICML-13), pages 1058–1066, 2013.

Sign up for DeepAI

Join one of the world's largest A.I. communities

Already have an account? Login here

Login to DeepAI

Don't have an account? Signup here

SHARE