what it was when this example was trained (since I just modified the default
value in dlib to something else).

code a little.

batch normalization running stats window size from 1000 to 100.

instead of rectangle.

attendant objects.  Also fixed a minor bug in the loss layer.

image.

that use std::vector in addition to dlib::array.

with the mapping functions necessary at each layer to support these routines.

visit_layers_range().

Add an overload of dnn_trainer::train_one_step that takes a pair of
iterators rather than a std::vector.

singular value of the input matrix rather than as an absolute tolerance.

calling code.

are empty, leading to MATLAB complaining about output arguments to being
assigned.