Average Sync Task
The Average sync task computes the arithmetic mean of a 3D tensor along a configurable axis on the GPU. Given an input tensor \(X \in \mathbb{C}^{B \times H \times W}\) (or \(\mathbb{R}^{B \times H \times W}\)), the task collapses the selected axis by averaging the elements between the indices start (inclusive) and end (exclusive):
where \(k\) is the axis being reduced.
Note
Axis labels follow the (B, H, W) layout: axis 0 averages across batches/frames, axis 1 across rows, and axis 2 across columns. The output retains the other two dimensions and replaces the reduced one with size 1.
Warning
For integer inputs (uint8/uint16), the averaged value is truncated when cast back to the original dtype.
Inputs
This task has a single input of shape (B, H, W), where:
B: number of frames or batchesH: height (rows)W: width (columns)
Valid dtypes are uint8, uint16, float32, and complex32. The tensor must reside in device memory.
Outputs
This task has a single output whose shape matches the input except that the reduced axis is set to 1:
- axis
0->(1, H, W) - axis
1->(B, 1, W) - axis
2->(B, H, 1)
The dtype of the output tensor matches the input dtype. The tensor is written to device memory.
Inplace
This task does not support inplace operation.
Ownership
This task does not own any inputs or outputs.
Settings
Example:
| Property | Pattern | Type | Deprecated | Definition | Title/Description |
|---|---|---|---|---|---|
| + axis | No | integer | No | - | Axis along which to compute the mean. Must satisfy: - axis in [0, 2] |
| + start | No | integer | No | - | Inclusive index where the average begins. Must satisfy: - start >= 0 - start < end |
| + end | No | integer | No | - | Exclusive index where the average stops. Must satisfy: - end > start - end <= size along axis |