To acquire color images, most commercial camerasrely on color filter arrays (CFAs), which are a pattern ofcolor filters overlaid over the sensor’s focal plane. Demosaickingdescribes the processing techniques to reconstruct a full colorimage for all pixels on the focal plane array. Most demosaickingmethods are tailored for a specific CFA, and tend to workpoorly for others. In this work we present an algorithm fordemosaicking a wide variety of CFAs. The proposed methodallows to blend the knowledge of the CFA with informationcoming from data, employing a novel transformation and pattern-invariant loss function. The method is based on the unrolling ofan algorithm based on a neural network learned on availableexamples. Preliminary experiments over RGB and RGBW CFAsshow that the method performs well over a range of CFAs and iscompetitive for CFAs for which competing methods were tailoredto work well on.