![]() |
My Project
|
|
For example, if the signal is two dimensional with m & n as sizes along the 0th & 1st dimensions respectively, then the possible batch operations are as follows.
Input Signal Dimensions | Filter Dimensions | Output Dimensions | Batch Mode | Explanation |
---|---|---|---|---|
[m n 1 1] | [m n 1 1] | [m n 1 1] | No Batch | Output will be a single convolve array |
[m n 1 1] | [m n p 1] | [m n p 1] | Filter is Batched | p filters applied to same input |
[m n p 1] | [m n 1 1] | [m n p 1] | Signal is Batched | 1 filter applied to p inputs |
[m n p 1] | [m n p 1] | [m n p 1] | Identical Batches | p filters applied to p inputs in one-to-one correspondence |
[m n p 1] | [m n 1 q] | [m n p q] | Non-overlapping batches | q filters applied to p inputs in to produce p x q results |
[m n 1 p] | [m n q 1] | [m n q p] | Non-overlapping batches | q filters applied to p inputs in to produce q x p results |