fn
adaptive_avg_pool2d_op
→TensorImplPtrint adaptive_avg_pool2d_op(const int & x, int OH, int OW)Python wrappers
Public Python APIs implemented by this engine symbol.Two-dimensional adaptive average-pool to a fixed (OH, OW) grid.
The canonical global average pooling head of modern classification
backbones (ResNet, EfficientNet, ConvNeXt, ...) uses
OH = OW = 1.
Parameters
xTensorImplPtrInput of shape
(B, C, H, W).OH, OWintTarget output extents. Must satisfy
H % OH == 0 and W % OW == 0.Returns
TensorImplPtrOutput of shape (B, C, OH, OW).
Raises
ShapeMismatchIf
x is not 4-D.not_implementedIf any spatial axis is not evenly divisible by its target.