An open source image recognition network using deep learning published by google, the repository is here. If there is an area similar to the pattern that was being learned in the input image, it will try to restore the original image and synthesize an image that is far from the human imagination. I think there are many people who have seen it because it is a topic recently.
You can do this locally! , Needs to install Caffe, which is a deep learning framework, so please install it. The introduction method of Caffe is summarized in this article, so please refer to it if you like.
It works with ipython notebook, so put it in.
pip install ipython "ipython[notebook]"
Clone the source published from the repository next to Caffe.
cd /path/to/caffe
cd ../
git clone [email protected]:google/deepdream.git
Since you need a trained caffe model, download it from caffemodel_url here and save it in models / bvlc_googlenet /
in Caffe.
Go to deepdream and launch ipython notebook.
cd deepdream
ipython trust dream.ipynb
ipython notebook
I think the browser has been launched. Select dream.ipynb and press the play button at the top. If you go to the end, deepdream will be executed based on the image sky1024px.jpg
, and the image converted todeepdream / frames /
will be saved.
With this, I was able to output the dream of the neural network (I say dream, but it's not a dream at all, I feel like I just forced it to recognize it), if you want to deepdream your favorite image, you can do it If you save it as sky1024px.jpg
, it will be loaded (because I don't want to mess with the contents, I wear it sideways).
When uploading the created image to SNS etc., add the #deepdream tag because the people who made deepdream will be pleased.
Recommended Posts