bulent2k2 / kojo-ai-2

AI in Kojo via Tensorflow-Java; and more...

Kojo-AI

Support for data science, machine learning, and more - within Kojo.

A previous cut of kojo-ai is available at https://github.com/litan/kojo-ai. This is the next cut, based on the official Tensorflow-Java API.

Current Features (work in progress, but functional):

• Neural Networks (via Tensorflow-Java).
• Plotting (via XCharts).
• Graphs (Bfs, Dfs, A* search, etc).
• Neural Style Transfer (NST) (via Tensorflow-Java and Arbitrary Style Transfer).
• Object Detection (via Tensorflow-Java).

Get Started

To use Kojo-AI within Kojo, just use a published release, or copy artifacts from this repo into the Kojo libk directory as per the following steps:
sbt buildDist
mv ~/.kojo/lite/libk ~/.kojo/lite/libk.bak
cp -var dist ~/.kojo/lite/libk

NST Example

An example showing NST in action is given below:

// #include /nst.kojo

cleari()
clearOutput()

val alpha = 0.8f
val fltr1 = new NeuralStyleFilter("/path/to/savedmodel/", "/path/to/style.jpg", alpha)
// val fltr2 = new NeuralStyleFilter("/path/to/savedmodel/", "/path/to/style2.jpg", alpha)

val drawing = Picture {
    setPenColor(cm.gray)
    var clr = cm.rgba(255, 0, 0, 127) // start with a semi transparent red color
    setPenThickness(8)
    repeat(18) {
        setFillColor(clr)
        repeat(5) {
            forward(100)
            right(72)
        }
        clr = clr.spin(360 / 18) // change color hue
        right(360 / 18)
    }
}

val pic = effect(fltr1) -> drawing
// or apply two filters
// val pic = effect(fltr1) * effect(fltr2) -> drawing

draw(pic)

NST sample images

Original Kojo Drawing:

Drawing with one NST filter applied:

Drawing with two NST filters applied: