feat_operator_convtranspose

Context

issue : aidge#235 (closed)

• ADDED :
  • feat : [ADD] PaddedConvTranspose operator
  • feat : [ADD] convtranspose operator
  • feat : added getDataType() function to OperatorTensor
  • feat : approxEqual, better error checking & error messages - replaced tensor->size by tensor->dims comparison - replaced fmt::print by log::error - replaced assert with AIDGE_ASSERT
• FIXES :
  • fix : Expand shape input was set as Param instead of Data
  • fix : missing header in Types.h
• CHORES :
  • chore : [DOC] added pybind documentation to graphview.compile
• Conv
  • feat : [Conv] added check to ensure dilation & stride attributes values are strictly positive
  • fix : [Conv] added check to ensure that dilation & stride values were all >= 1
    • better warning message for conv operator
  • chore : [Conv] updated the method for attribute accession to ease code reading
  • fix : [Conv] added check to ensure that dilation & stride values were all >= 1 & better warning message for conv operator constructor
  • chore : [Conv] updated the method for attribute accession to ease code reading

src/operator/Conv.cpp

@@ -105,18 +105,18 @@ Aidge::Conv_Op<DIM>::computeReceptiveField(
         std::vector<DimSize_t> inputDims{outputDims[0], getInput(0)->dims()[1]};
         for (DimIdx_t i = 0; i < DIM; ++i) {
             inputDims.push_back((outputDims[2+static_cast<std::size_t>(i)] - 1)
-                        * mAttributes->template getAttr<ConvAttr::StrideDims>()[static_cast<std::size_t>(i)]
+                        * strideDims()[static_cast<std::size_t>(i)]
                         + 1
-                        + (mAttributes->template getAttr<ConvAttr::KernelDims>()[static_cast<std::size_t>(i)] - 1)
-                        * mAttributes->template getAttr<ConvAttr::DilationDims>()[static_cast<std::size_t>(i)]);
-            inputIdxDims[2+i] *= mAttributes->template getAttr<ConvAttr::StrideDims>()[static_cast<std::size_t>(i)];
+                        + (kernelDims()[static_cast<std::size_t>(i)] - 1)
+                        * dilationDims()[static_cast<std::size_t>(i)]);
+            inputIdxDims[2+i] *= strideDims()[static_cast<std::size_t>(i)];
         }
 
         // Weight
         // same output value, every input channel is used
         std::vector<DimSize_t> weightDims{outputDims[1], getInput(0)->dims()[1]};
         for (std::size_t i = 0; i < DIM; ++i) {
-            weightDims.push_back(mAttributes->template getAttr<ConvAttr::KernelDims>()[i]);
+            weightDims.push_back(kernelDims()[i]);
         }
         std::vector<DimSize_t> weightIdxDims = std::vector<DimSize_t>(DIM+2, 0);
         weightIdxDims[0] = firstEltDims[1];