Known ML Problems

The Min/Max Problem

The Problem:
Page processing of images implies that the entire image is usually not completely processed. So it is not possible to calculate some properties of the result image correctly, e.g. the minimum or maximum values. These values, however, are part of the image properties which must be specified when describing the output image properties of modules. So modules must not rely on the correctness of the min-max interval; they can assume that the interval includes all voxel values, but it may be larger than the real values.

Possible solutions:
The normal way to solve this problem is to simply estimate the minimum and maximum values from the algorithm of the module and the min/max values of the input images. This normally leads to larger min/max intervals than necessary, which, however, is explicitly allowed. Only for a few algorithms this is difficult or impossible; there typically the range of the used voxel data type is used.

The Notification-Consistency Problem

Notification between ML modules and fields are heavily used in module networks. It has become one of the most powerful features of the ML and provides a mechanism to implement much algorithmic behavior in connections instead of in additional program code managing the module network. Especially in MeVisLab this feature is strongly used also in connection with macro and OpenInventor modules.

The notification mechanism running within the ML to manage this has its limitations; it cannot foresee all possible module connections, field observers, sensors, macros, direct and indirect notifications resulting from additional code controlling the module network etc. So you can possibly construct image, module and field connections which lead to incomplete updates or to undesired network behavior.
If you run into such a situation please report the scenario so that we can try to fix that problem.