fn
adaptive_avg_pool3d_op
→TensorImplPtrint adaptive_avg_pool3d_op(const int & x, int OD, int OH, int OW)Python wrappers
Public Python APIs implemented by this engine symbol.Three-dimensional adaptive average-pool to a fixed (OD, OH, OW) volume.
Used as the final pooling layer in 3-D classification networks
(3D-ResNet, SlowFast). OD = OH = OW = 1 yields the volumetric
global-average-pool head.
Parameters
xTensorImplPtrInput of shape
(B, C, D, H, W).OD, OH, OWintTarget output extents. Each must evenly divide the corresponding input axis.
Returns
TensorImplPtrOutput of shape (B, C, OD, OH, OW).
Raises
ShapeMismatchIf
x is not 5-D.not_implementedIf any spatial axis is not evenly divisible by its target.