|
|
@ -2307,8 +2307,8 @@ class Conv2DTranspose(layers.Layer):
|
|
|
|
filter, and dilations, strides, paddings. Input and output
|
|
|
|
filter, and dilations, strides, paddings. Input and output
|
|
|
|
are in NCHW format. Where N is batch size, C is the number of feature map,
|
|
|
|
are in NCHW format. Where N is batch size, C is the number of feature map,
|
|
|
|
H is the height of the feature map, and W is the width of the feature map.
|
|
|
|
H is the height of the feature map, and W is the width of the feature map.
|
|
|
|
Filter's shape is [MCHW] , where M is the number of output feature map,
|
|
|
|
Filter's shape is [MCHW] , where M is the number of input feature map,
|
|
|
|
C is the number of input feature map, H is the height of the filter,
|
|
|
|
C is the number of output feature map, H is the height of the filter,
|
|
|
|
and W is the width of the filter. If the groups is greater than 1,
|
|
|
|
and W is the width of the filter. If the groups is greater than 1,
|
|
|
|
C will equal the number of input feature map divided by the groups.
|
|
|
|
C will equal the number of input feature map divided by the groups.
|
|
|
|
If bias attribution and activation type are provided, bias is added to
|
|
|
|
If bias attribution and activation type are provided, bias is added to
|
|
|
|