This class implements a layer which performs max pooling on the set of sequences, with the filter applied on BatchLength
axis.
void SetFilterLength(int length);
Sets the filter length. If you set the length to 0
or less, the maximum over the whole axis will be calculated.
void SetStrideLength(int length);
Sets the filter stride. This value is meaningful only if the filter length is greater than 0
.
The layer has no trainable parameters.
The single input accepts a blob of the following dimensions:
BatchLength
- the sequence lengthBatchWidth * ListSize
- the number of sequences in the setHeight * Width * Depth * Channels
- the length of each vector in the sequences
The single output contains a blob with the results.
If the filter length is positive, the BatchLength
of the resulting blob may be calculated as
(BatchLength - GetFilterLength()) / GetStrideLength() + 1
.
If the filter length is equal or less than 0
, the BatchLength
of the resulting blob is 1
.
The rest of the dimensions of the output blob are equal to the corresponding dimensions of the input.