[English]
This example shows how to classify fashion item images (fashion MNIST) and to train a conditional generative adversarial network (CGAN) to generate fashion item images.
This demo was created based on the Matlab official document entitled Train Conditional Generative Adversarial Network (CGAN) [1].
Conditinal Generative Adversarial Network (CGAN) was proposed in the paper [2]. This livescript was made with the references such as [3].
A open dataset called fashion MNIST was used in this demo [4]. A helper function to process fashion MNIST data was created in the official document [5].
This demo is a modified version of the file entitled 'Conditional GAN with MNIST' [6].

[Japanese]
このデモでは、１）深層学習を用いた、fashion MNISTデータの分類や、２）conditional GAN (CGAN)を用いたfashion MNISTデータの生成を行います。ここでは、公式ドキュメント [1] を参考にしました。また、conditional GANの論文は [2]です。生成結果はサムネイルのgif画像になります。

[Key Words]
adversarial generative network, CGAN, classification, cnn, conditional GAN, deep learning, fashion MNIST, GAN, MNIST

[1] https://jp.mathworks.com/help/deeplearning/ug/train-conditional-generative-adversarial-network.html]
[2] Mirza M, Osindero S. Conditional generative adversarial nets. arXiv preprint arXiv:1411.1784. 2014 Nov 6.
https://arxiv.org/pdf/1411.1784.pdf
[3] 山下隆義、イラストで学ぶ　ディープラーニング　改訂第2版、ＫＳ情報科学専門書、2018年
[4] Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms. Han Xiao, Kashif Rasul, Roland Vollgraf. arXiv:1708.07747
[5] Matlab official document: Visualize High-Dimensional Data Using t-SNE
[6] Conditional GAN with MNIST (https://jp.mathworks.com/matlabcentral/fileexchange/74921-conditional-gan-generative-adversarial-network-with-mnist)