https://github.com/primaryobjects/fashion
The Fashion-MNIST dataset and machine learning models.
https://github.com/primaryobjects/fashion
artificial-intelligence artificial-neural-networks classification data-science dataset fashion fashion-mnist image-classification image-recognition machine-learning mnist r supervised-learning support-vector-machines svm xgboost
Last synced: 7 months ago
JSON representation
The Fashion-MNIST dataset and machine learning models.
- Host: GitHub
- URL: https://github.com/primaryobjects/fashion
- Owner: primaryobjects
- Created: 2017-09-15T18:15:16.000Z (about 8 years ago)
- Default Branch: master
- Last Pushed: 2017-10-24T13:34:00.000Z (almost 8 years ago)
- Last Synced: 2025-03-21T06:41:37.845Z (7 months ago)
- Topics: artificial-intelligence, artificial-neural-networks, classification, data-science, dataset, fashion, fashion-mnist, image-classification, image-recognition, machine-learning, mnist, r, supervised-learning, support-vector-machines, svm, xgboost
- Language: R
- Size: 28.8 MB
- Stars: 24
- Watchers: 4
- Forks: 8
- Open Issues: 0
-
Metadata Files:
- Readme: readme.md
Awesome Lists containing this project
README
Fashion
=======
Training AI machine learning models on the Fashion MNIST [dataset](https://github.com/zalandoresearch/fashion-mnist).
Read the full article at [Image Recognition for Fashion with Machine Learning](http://www.primaryobjects.com/2017/10/23/image-recognition-for-fashion-with-machine-learning/)
## What is Fasion-MNIST?
Fashion-MNIST is a dataset consisting of 70,000 images (60k training and 10k test) of clothing objects, such as shirts, pants, shoes, and more. Each example is a 28x28 grayscale image, associated with a label from 10 classes. The 10 classes are listed below.
The dataset was designed as a plug-and-play replacement for the traditional [MNIST](http://yann.lecun.com/exdb/mnist/) handwritten digit recognizing dataset. Both datasets use the same scale and type of image files, along with the same number of classification labels (10), for usage with machine learning models.
## The Dataset
Download the dataset as a series of 4 zipped files:
[Train Images](/data/train-images-idx3-ubyte.gz)
[Train Labels](/data/train-labels-idx1-ubyte.gz)
[Test Images](/data/t10k-images-idx3-ubyte.gz)
[Test Labels](/data/t10k-labels-idx1-ubyte.gz)
The dataset can be loaded in R by using the same [script](https://gist.github.com/brendano/39760) for loading the traditional MNIST dataset. Additionally, images can be displayed in the same manner, by calling the `showDigit` method.
### Labels
Each training and test example is assigned to one of the following labels:
| Label | Description |
| --- | --- |
| 0 | T-shirt/top |
| 1 | Trouser |
| 2 | Pullover |
| 3 | Dress |
| 4 | Coat |
| 5 | Sandal |
| 6 | Shirt |
| 7 | Sneaker |
| 8 | Bag |
| 9 | Ankle boot |
## Pre-trained Models
You can find a set of [pre-trained models](https://github.com/primaryobjects/fashion/tree/master/data/models) included in this repository for immediate loading into the R environment.
## Loading Your Own Images
You can [load](https://gist.github.com/primaryobjects/06c2deca989af9c1acf735521ba9db81#file-readpng2-r) your own (color) images of shirts, pants, or shoes by reading any 28x28 image, converting to grayscale, and extracting the single channel of bytes from grayscale. Once loaded, you can classify the result using a trained machine learning model, based on the fashion-mnist dataset.
## Accuracy
The trained models have achieved the following accuracies (train/test) on 30k images as shown below. A detailed list of accuracies is also [available](https://github.com/zalandoresearch/fashion-mnist#benchmark).
#### Baseline Algorithm (always predicts the most frequently occurring clothing item)
12%/10%
#### Support Vector Machine (svmRadial)
91%/87%
#### Gradient Boosting Machine (gbm)
90%/85%
#### Neural Network (multinom)
84%/78%
#### LogitBoost
81%/75%
## References
[The MNIST Database of Handwritten Digits](http://yann.lecun.com/exdb/mnist/)
[The Fashion-MNIST Database](https://github.com/zalandoresearch/fashion-mnist)
[Zalando](https://jobs.zalando.com/tech/)
## License
MIT
## Author
Kory Becker
http://www.primaryobjects.com