A GAN approach for generating handwritten digits with a deep neural network written in Keras.
In both notebooks, the MNIST dataset is used. A generative adversarial network (GAN) is deployed to create unique images of handwritten digits. The generated images look like they're taken from the dataset (that is the purpose), but they are generated from scratch (actually, from noise) and are all unique.
In GAN-keras-mnist-MLP.ipynb, a multilayer perceptron network is used for the generator and the discriminator In GAN-keras-mnist-DCGAN.ipynb, a transposed convolutional (or deconvolutional) network is used for the generator and a regular convolutional network is used for the discriminator.
Otherwise, the approach is kept the same
You can see some generated samples within each of the notebooks
Further experimentation is required.
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。