concept DeepDream in category deep learning

This is an excerpt from Manning's book Deep Learning with R.
DeepDream is an artistic image-modification technique that uses the representations learned by convolutional neural networks. It was first released by Google in the summer of 2015, as an implementation written using the Caffe deep-learning library (this was several months before the first public release of TensorFlow).[4] It quickly became an internet sensation thanks to the trippy pictures it could generate (see, for example, figure 8.3), full of algorithmic pareidolia artifacts, bird feathers, and dog eyes—a byproduct of the fact that the DeepDream convnet was trained on ImageNet, where dog breeds and bird species are vastly overrepresented.
You’ll start from a convnet pretrained on ImageNet. In Keras, many such convnets are available: VGG16, VGG19, Xception, ResNet50, and so on. You can implement DeepDream with any of them, but your convnet of choice will naturally affect your visualizations, because different convnet architectures result in different learned features. The convnet used in the original DeepDream release was an Inception model, and in practice Inception is known to produce nice-looking DeepDreams, so you’ll use the Inception V3 model that comes with Keras.

This is an excerpt from Manning's book Deep Learning with Python.
DeepDream is an artistic image-modification technique that uses the representations learned by convolutional neural networks. It was first released by Google in the summer of 2015, as an implementation written using the Caffe deep-learning library (this was several months before the first public release of TensorFlow).[4] It quickly became an internet sensation thanks to the trippy pictures it could generate (see, for example, figure 8.3), full of algorithmic pareidolia artifacts, bird feathers, and dog eyes—a byproduct of the fact that the DeepDream convnet was trained on ImageNet, where dog breeds and bird species are vastly overrepresented.
You’ll start from a convnet pretrained on ImageNet. In Keras, many such convnets are available: VGG16, VGG19, Xception, ResNet50, and so on. You can implement DeepDream with any of them, but your convnet of choice will naturally affect your visualizations, because different convnet architectures result in different learned features. The convnet used in the original DeepDream release was an Inception model, and in practice Inception is known to produce nice-looking DeepDreams, so you’ll use the Inception V3 model that comes with Keras.