DeepDream through all the layers of GoogleNet

Example of the generated images using DeepTrip [1].
DeepTrip is a modification of the DeepDream code [2] to execute various steps of the code at every layer of a Convolutional Neural Network. The initial purpose was to visualize the type of images that are maximized at every layer.
Other parts of the code have been adapted from DeepDreamVideo [3].

The video has been rendered using the program SlowmoVideo [4]. The program generated some rendering artifacts in some frames. It is possible to remove them but it requires additonal time.

More information and a set of best frames per layer can be found in my Aalto University personal webpage: https://users.ics.aalto.fi/perellm1/deep_dreams.shtml

[1] https://github.com/perellonieto/DeepTrip
[2] https://github.com/google/deepdream
[3] https://github.com/graphific/DeepDreamVideo
[4] http://slowmovideo.granjow.net/

Видео DeepDream through all the layers of GoogleNet канала Miquel Perelló Nieto