Extensions were applied to the code which deals with MDL-based shape model construction (Chapter cha:MDL Models).
As an example, below lies pseudo-code of what is referred to as adaptive precision. It simply selects a suitable tolerance for the optimiser. It does so in order to avoid excessive computations at early stages of the landmark selection procedure.
INPUTS: iterations_ratio, precision_automation_type
% iterations_ratio is the ratio between the
% current iteration and the total number
% of iterations
switch precision_automation_type,
case 'default'
if (iterations_ratio < 0.1),
return precision_required = 1e-1;
elseif (iterations_ratio < 0.2),
return precision_required = 1e-2;
elseif (iterations_ratio < 0.3),
return precision_required = 1e-3;
...
else
return precision_required = 1e-12;
end
case 'smart'
% Look at the evaluation curve derivatives to
% infer precision
end