Grouped convolution applies a separate filter over each input channel To check for a grouped convolution we can use the following calculation: numGroups = input[channels] / filter[in_channels] This gives us the number of groups to divide the input channel into. An assumption is made here that numGroups > 1 implies a grouped convolution We do not support grouped convolution, therefore we will retur
| 331 | // An assumption is made here that numGroups > 1 implies a grouped convolution |
| 332 | // We do not support grouped convolution, therefore we will return unsupported if numGroups > 1 |
| 333 | bool IsGroupedConvolution(armnn::TensorShape inputShape, |
| 334 | armnn::TensorShape filterShape, |
| 335 | const armnn::DataLayout dataLayout) |
| 336 | { |
| 337 | const armnnUtils::DataLayoutIndexed dataLayoutIndexed(dataLayout); |
| 338 | const unsigned int channelsIndex = dataLayoutIndexed.GetChannelsIndex(); |
| 339 | return inputShape[channelsIndex] / filterShape[channelsIndex] > 1; |
| 340 | } |
| 341 | |
| 342 | // Function that takes a TensorInfo Parameter and returns the same TensorInfo with data type FLoat32. |
| 343 | armnn::TensorInfo ConvertTensorInfoToFloat32(const armnn::TensorInfo& tensorInfo) |
no test coverage detected