Conv2DTransposeBatchnormFusionPass

Description

Fuses the Conv2DTranspose+BatchNorm/BNInference operators into the Conv2DTransposeD operator, that is, converts the input of BatchNorm or BNInference into the weight and bias inputs of Conv2DTransposeD.

Restrictions

• The groups attribute of Conv2DTranspose must be set to 1.
• The input type (except input x) of BatchNorm or BNInference must be const.
• Only static scenarios are supported.
• The BatchNorm operator supports only four dimensions.