Skip to content

Commit

Permalink
elaborate more on CNN.
Browse files Browse the repository at this point in the history
  • Loading branch information
Julio Jerez committed Nov 6, 2023
1 parent e2f3a33 commit ccccbff
Showing 1 changed file with 27 additions and 27 deletions.
54 changes: 27 additions & 27 deletions newton-4.00/sdk/dBrain/ndBrainLayerConvolutional.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -665,13 +665,13 @@ void ndBrainLayerConvolutional::MakePrediction(const ndBrainVector& input, ndBra

auto PredictOutputChannel = [this](const ndBrainVector& input, const ndBrainVector& kernels, ndBrainFloat bias, ndBrainVector& output)
{
auto CrossCorrelation = [this](const ndBrainVector& input, const ndBrainVector& kernels)
auto CrossCorrelation = [this](const ndBrainVector& input, const ndBrainVector& kernel)
{
ndBrainFloat value = ndBrainFloat(0.0f);
for (ndInt32 i = 0; i < m_inputOffsets.GetCount(); ++i)
for (ndInt32 i = m_inputOffsets.GetCount() - 1; i >= 0; --i)
{
ndInt32 index = m_inputOffsets[i];
value += input[index] * kernels[i];
value += input[index] * kernel[i];
}
return value;
};
Expand Down Expand Up @@ -801,7 +801,6 @@ void ndBrainLayerConvolutional::CalculateParamGradients(
m_paddedGradientOut.Set(ndBrainFloat(0.0f));
}


ndInt32 inputOffset = 0;
ndBrainFloat convKernelBuffer[256];
ndBrainMemVector convKernel(convKernelBuffer, kernelSize);
Expand Down Expand Up @@ -834,38 +833,39 @@ void ndBrainLayerConvolutional::CalculateParamGradients(

auto CalculateInpuGradient = [this](const ndBrainVector& filter, ndBrainVector& output)
{
//auto CrossCorrelation = [this](const ndBrainVector& input, const ndBrainVector& kernels)
//{
// ndBrainFloat value = ndBrainFloat(0.0f);
// for (ndInt32 i = 0; i < m_inputOffsets.GetCount(); ++i)
// {
// ndInt32 index = m_inputOffsets[i];
// value += input[index] * kernels[i];
// }
// return value;
//};
auto CrossCorrelation = [this, &filter](const ndBrainVector& input)
{
ndBrainFloat value = ndBrainFloat(0.0f);
for (ndInt32 i = m_inputGradOffsets.GetCount() - 1; i >= 0 ; --i)
{
ndInt32 index = m_inputGradOffsets[i];
value += input[index] * filter[i];
}
return value;
};

ndInt32 outputOffset = 0;
ndInt32 gradInputOffset = 0;
const ndInt32 gradInputWidth = m_inputWidth + m_kernelSize - 1;
const ndInt32 gradInputSize = gradInputWidth * m_kernelSize;
//const ndInt32 inputSize = m_inputWidth * m_kernelSize;
//const ndInt32 kernelSize = m_kernelSize * m_kernelSize;
const ndInt32 kernelSize = m_kernelSize * m_kernelSize;
for (ndInt32 y = 0; y < m_inputHeight; ++y)
{
for (ndInt32 x = 0; x < m_inputWidth; ++x)
{
ndBrainFloat value = ndBrainFloat(0.0f);
// const ndBrainMemVector in(&input[inputOffset + i], inputSize);
//
// ndInt32 kernelOffset = 0;
// for (ndInt32 k = 0; k < m_inputDepth; ++k)
// {
// const ndBrainMemVector filter(&kernels[kernelOffset], kernelSize);
// value += CrossCorrelation(in, filter);
// }
// kernelOffset += kernelSize;
// //ndAssert(value == output[outputOffset + i]);
output[outputOffset + x] = value;
//ndBrainFloat value = ndBrainFloat(0.0f);
//const ndBrainMemVector in(&input[inputOffset + i], inputSize);
const ndBrainMemVector in(&m_paddedGradientOut[gradInputOffset + x], gradInputSize);

//ndInt32 kernelOffset = 0;
//for (ndInt32 k = 0; k < m_inputDepth; ++k)
//{
// const ndBrainMemVector filter(&kernels[kernelOffset], kernelSize);
// value += CrossCorrelation(in, filter);
//}
//kernelOffset += kernelSize;
output[outputOffset + x] += CrossCorrelation(in);
}
outputOffset += m_inputWidth;
gradInputOffset += gradInputWidth;
Expand Down

0 comments on commit ccccbff

Please sign in to comment.