sheerun / neural-style-server

neural-style-server

A web server frontend for https://github.com/jcjohnson/neural-style

Setup

Dependencies:

• jcjohnson/neural-style
• nodejs (tested with v0.10.25)

Follow the neural-style setup instructions for installing neural-style and all of its dependencies. neural-style-server assumes that neural-style is being used with NVidia hardware, CUDA, and CuDNN (e.g. it depends on nvidia-smi for enumerating GPUs).

Once you have neural-style set up, clone the neural-style-server repository. You'll need to edit config/default.json or add your own config file e.g. config/local.json - see the node-config documentation for more information about config file naming and load order.

Config options:

• port: (required) The HTTP server port on which the web server will listen.
• dataPath: (required) The directory where image and settings input files and image output files will be stored. There shouldn't be any other files in this directory - neural-style-server will scan it on startup to discover data created by previous runs.
• neuralStylePath: (requierd) The directory where the neural-style repository is cloned.
• username: (optional) A username to use for HTTP basic auth. May be omitted to disable auth.
• password: (optional) A password to use for HTTP basic auth. May be omitted to disable auth.

Run

You must run neural-style-server from a shell that has environment variables configured to use Torch (if you did not allow the Torch installer to configure this by default, you can run source install/bin/torch-activate from the Torch directory to set up the necessary environment).

To launch neural-style-server, run:

nodejs app.js

Screenshot