Design

Images

Unsurprisingly, the overwhelming majority of operators in Imagecat either generate, modify, or consume images, making them one of the most important datatypes in the library.

In Imagecat, an image is a container for a set of named layers, and a layer contains one or more channels. For example, an image loaded by the imagecat.load operator might contain a layer named “C” with three channels of RGB color data, and a layer “A” with a single channel of opacity information. More sophisticated workflows typical of 3D computer graphics might involve images containing dozens of layers containing color, depth, lighting, normal, object identity, velocity information, and more.

To accomodate this, every image in Imagecat is an instance of imagecat.data.Image containing a dict of str layer names as keys, and instances of imagecat.data.Layer containing the layer data as values. Any string is a valid layer name, although some names are more common by convention, such as “C” for RGB color information and “A” for alpha channels. Layer data is always stored using Numpy arrays, which always have three dimensions, in (row, column, channel) order. Most layer data contains half-precision floating point values with linear brightness, instead of the gamma-adjusted unsigned bytes with which you may be familiar. Note that the third dimension is present even for single-channel layers like alpha channels and masks, in which case its size will be one. When working with layers, be careful to keep in mind that (row, column) slicing is the opposite of the (x, y) ordering normally used in image processing. Finally, each layer has a imagecat.data.Role, which defines how the data in the layer will be used. For example, a layer with three channels might contain RGB color information, or XYZ coordinates in world space, or XYZ velocity vectors, or UVW coordinates in texture space. The layer role is what distinguishes among these cases, which is useful in visualization and I/O.

Graph

Imagecat operators are designed to be used with Graphcat, https://graphcat.readthedocs.io, with which they are organized into computational graphs that manage executing them at the right time and in the right order. Imagecat operators are compatible with all graphcat.graph.Graph derivatives, which include graphcat.StaticGraph, graphcat.DynamicGraph, and graphcat.StreamingGraph.

Named Inputs

Named inputs are the objects that Graphcat uses to pass inputs from one computational graph task to another; depending on the type of Graph that you’re using, your Imagecat operators will be called with one of graphcat.static.NamedInputs, graphcat.dynamic.NamedInputs, or graphcat.streaming.NamedInputs, which all have compatible APIs.

Note

Since every Graphcat task has the same parameters, the reference documentation for each Imagecat operator lists the named input parameters in a separate “Named Inputs” section.