Skip to content

Implementation of the DeepDream computer vision algorithm in TensorFlow, following the guide on the TensorFlow repo.

License

Notifications You must be signed in to change notification settings

EdCo95/deepdream

Repository files navigation

DeepDream in TensorFlow

DeepDream is a computer vision algorithm which produces psychedelic images by enhancing patterns in existing images using a deep neural network. It works by performing gradient ascent with regard to a particular neuron so that that neuron becomes more confident of its classification as it alters the image. This helps to visualise what a network has learned, but more importantly the images it produces are awesome.

DeepDream was invented by Google, applied to the Inception network developed for ImageNet in 2014. The original algorithm is in Caffe. This repo is a TensorFlow implementation. It is, essentailly, exactly the same as the TensorFlow guide. All of the important code is identical, however rather than running each of the increasingly complex algorithms in an IPython Notebook, I have instead created separate code files for each of the algorithms so that each can be run in isolation - hopefully preventing confusion over exactly what is needed for each algorithm.

I have added some comments to try and wrap my head around what is going on, but I am not creating anything new here, merely trying to understand an algorithm that produces interesting results.

References

About

Implementation of the DeepDream computer vision algorithm in TensorFlow, following the guide on the TensorFlow repo.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages