{"id":3482,"date":"2016-04-21T14:39:22","date_gmt":"2016-04-21T05:39:22","guid":{"rendered":"https:\/\/deeplearning.jp\/?page_id=693"},"modified":"2017-07-21T13:00:01","modified_gmt":"2017-07-21T04:00:01","slug":"cvae","status":"publish","type":"page","link":"https:\/\/deeplearning.jp\/en\/cvae\/","title":{"rendered":"Conditional Variational Autoencoder (CVAE)"},"content":{"rendered":"

We would like to introduce conditional variational autoencoder (CVAE) [2][3], a deep generative model,\u00a0and show our online demonstration (Facial VAE<\/a>).\u00a0We believe that the CVAE method is very promising to many fields, such as image generation, anomaly detection problems, and so on.<\/p>\n

VAE<\/h2>\n

Variational autoencoder (VAE) [1] is a generative model which utilizes deep neural networks to describe the distribution of observed and latent (unobserved) variables. Using the VAE model,\u00a0we assume that the data x is generated by p\u03b8(x|z) where \u03b8 denotes the parameter of deep neural networks. Given the data x, we want to maximize the log-likelihood logp(x)\u00a0and in order to solve this problem, we can use variational inference. The main principle of variatioal inference is to introduce an approximate distribution q(z|x) and maximize the lower bound instead of the log-likelihood. In VAE, the approximate distribution q(z|x) can be parameterized by q\u03c6(z|x) where \u03c6 means the parameter of deep neural networks. We usually call p(x|z) the encoder (or the recognition model) and q(z|x) the decoder (or the generative model). From this point of view, we can consider this model as a different kind of autoencoder and why it is called variational \u201cautoencoder\u201d.\u00a0By\u00a0using the training algorithm Stochastic Gradient Variational Bayes (SGVB),\u00a0<\/i>we can train deep neural networks on VAE by backpropagation. If you want to know VAE algorithm in detail, please see [1].<\/p>\n

VAE can reconstruct the data such as normal autoencoders. Moreover, VAE can generate samples from random values by using the decoder p(x|z). Note that these generated samples are not actually exist in the original training data. VAE trains the probability distribution p(x|z) which is most likely to generate the original data, therefore we can generate new samples which look like the original data.<\/p>\n

\"vae\"<\/a><\/p>\n

CVAE<\/h2>\n

Conditional VAE [2][3] is VAE architecture conditioning on another description of the data, y.\u00a0 In this model, we can generate samples from the conditional distribution p(x|y). By changing the value of y, such as numbers of labels in MNIST, we can get corresponding samples x~p(x|y).<\/p>\n

\"cvae\"<\/a><\/p>\n

Application : Facial VAE<\/h2>\n

To demonstrate and explain how CVAE can apply to real face images we created \u201cFacial VAE\u201d.\u00a0This demo is inspired from [4][5]. In our demo, we can generate face images conditioned on attributes, such as male, young and so on. Moreover, we combine VAE and Generative Adversarial Networks (GAN) [6]\u00a0 into one model in order to generate clearer images.\u00a0 The most significant difference between [4][5] and ours\u00a0is the ability to infer attributes from an image given by a user.\u00a0Therefore, if you submit your image to Facial VAE, you will get the corresponding attributes and a reconstruction of the image.\u00a0 Moreover, if you change the attributes, then you will get a new generated image conditioned based on those attributes.<\/p>\n

\"overview\"<\/a><\/p>\n

Facial VAE –\u00a0Demo<\/a><\/p>\n

Please give try our demo!<\/b><\/p>\n

[1] Kingma, Diederik P., and Max Welling. “Auto-encoding variational bayes.” arXiv preprint arXiv:1312.6114<\/i> (2013).<\/p>\n

[2] Kingma, Diederik P., et al. “Semi-supervised learning with deep generative models.” Advances in Neural Information Processing Systems<\/i>. 2014.<\/p>\n

[3] Sohn, Kihyuk, Honglak Lee, and Xinchen Yan. “Learning Structured Output Representation using Deep Conditional Generative Models.” Advances in Neural Information Processing Systems<\/i>. 2015.<\/p>\n

[4] Larsen, Anders Boesen Lindbo, S\u00f8ren Kaae S\u00f8nderby, and Ole Winther. “Autoencoding beyond pixels using a learned similarity metric.” arXiv preprint arXiv:1512.09300<\/i> (2015).<\/p>\n

[5] Generating Faces with Torch (http:\/\/torch.ch\/blog\/2015\/11\/13\/gan.html<\/a>)<\/p>\n

[6] Goodfellow, Ian, et al. “Generative adversarial nets.” Advances in Neural Information Processing Systems<\/i>. 2014.<\/p>","protected":false},"excerpt":{"rendered":"

We would like to introduce conditional variational autoencoder (CVAE) [2][3], a deep more <\/i><\/a><\/p>\n","protected":false},"author":5,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"_acf_changed":false,"footnotes":""},"tags":[],"class_list":["post-3482","page","type-page","status-publish","hentry"],"acf":[],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\t\n\n\t\t\n