Adding New Transforms

This guide explains how to implement and register new transforms in the TransformsV2 system.

1. Implement the Transform Function

A transform is simply a function (or a callable object). It should be stateless and pure if possible.

Element-Wise Transform

For operations that transform a single element (e.g., a point, a mask, a number).

// Example: Scale a point by a factor
struct ScalePointParams {
    float factor = 1.0f;
};

Point2D ScalePoint(Point2D const& input, ScalePointParams const& params) {
    return {input.x * params.factor, input.y * params.factor};
}

Time-Grouped Transform

For operations that need access to all elements at a single time point (e.g., filtering, reduction).

// Example: Sum of a vector of floats
float SumReduction(std::span<float const> inputs) {
    float sum = 0.0f;
    for (float v : inputs) sum += v;
    return sum;
}

2. Registration

// In RegisteredTransforms.cpp
registry.registerTransform<Point2D, Point2D, ScalePointParams>(
    "ScalePoint",
    ScalePoint,
    "Scales a point by a factor."
);

Metadata

The registry automatically extracts type information. You can also provide:

Description: A human-readable description.
Category: To group transforms in the UI.
Keywords: For search.

3. Parameter Structs

If your transform takes parameters, define a simple struct.

Reflection: You must provide reflection for your parameter struct so the UI can generate a configuration form.
Serialization: Ensure the struct is serializable to/from JSON.

struct ScalePointParams {
    float factor = 1.0f;
    
    // Macro for reflection (example)
    REFLECT(ScalePointParams, factor)
};