| MPSCNNUpsamplingGradient(3) | MetalPerformanceShaders.framework | MPSCNNUpsamplingGradient(3) |
MPSCNNUpsamplingGradient
#import <MPSCNNUpsampling.h>
Inherits MPSCNNGradientKernel.
Inherited by MPSCNNUpsamplingBilinearGradient, and MPSCNNUpsamplingNearestGradient.
(nonnull instancetype) - initWithDevice:
double scaleFactorX
double scaleFactorY
This depends on Metal.framework The MPSCNNUpsamplingGradient filter is used for training. It is the backward filter for the MPSCNNUpsampling filter. It operates on the gradient input, specifically, it reduces the size of the gradient input in the x and y dimensions.
The number of output feature channels remains the same as the number of input feature channels.
The scaleFactor must be an integer value >= 1. The default value is 1. If scaleFactor == 1, the filter acts as a copy kernel.
Nearest and bilinear variants are supported.
For example, for the nearest variant with scaleFactorX = scaleFactorY = 2, the forward pass produced the following output:
Input: Output: a a b b a b a a b b c d c c d d c c d d
To upsample the image, the input data is replicated.
And, the backward pass for the above froward pass is computed in the following way:
Input: Output: a1 a2 b1 b2 a2 a3 b3 b4 x y c1 c2 d1 d2 z w c3 c4 d3 d4
where x = a1 + a2 + a3 + a4 y = b1 + b2 + b3 + b4 z = c1 + c2 + c3 + c4 w = d1 + d2 + d3 + d4
Standard init with default properties per filter type
Parameters:
Returns:
Reimplemented from MPSCNNGradientKernel.
The downsampling scale factor for the x dimension. The default value is 1.
The downsampling scale factor for the y dimension. The default value is 1.
Generated automatically by Doxygen for MetalPerformanceShaders.framework from the source code.
| Mon Jul 9 2018 | Version MetalPerformanceShaders-119.3 |