DeepBoof is a Java library for running deep neural networks trained using other projects (e.g. Torch and Caffe) with a focus on processing image data. Additional tools include visualization and network training. Image processing is done using BoofCV. While it has been designed to work with Torch and Caffe it does not depend either library for its core functionality.
Gradle is used to build DeepBoof and will automatically download all dependencies.
To get started classifying images simply load the Gradle project into you favorite IDE (probably IntelliJ or Eclipse) and run ExampleClassifyVggCifar10. This example downloads the network and testing data, then starts the classifier.
Alternatively, you can just build and run any example from the command-line using Gradle directly:
gradle exampleRun -Pwhich=ExampleClassifyCifar10TestSetDeepBoof is in an early state of development. The following are areas it needs the most improvement in:
- Support for Caffe models (already started)
- Native implementations of functions
- GPU implementations of functions