This class implements a layer that finds the optimal class sequence using the output of the CCrfLayer.
The layer has no settings.
The layer has no trainable parameters.
The layer has two inputs:
The first input accepts a blob with int
data that contains the optimal class sequences (i.e. first output of the CCrfLayer). The dimensions are:
BatchLength
is the sequence lengthBatchWidth
is the number of sequences in the setChannels
is equal to the number of classes- the rest of the dimensions are equal to
1
The second input accepts a blob with float
data that contains non-normalized logarithm of optimal class sequences probabilities (i.e. the second output of the CCrfLayer). It has the same dimensions as the first input.
The single output is a blob with int
data that contains the optimal class sequences. The blob dimensions are:
BatchLength
is equal to the inputs'BatchLength
BatchWidth
is equal to the inputs'BatchWidth
- the other dimensions are equal to
1