microsoft
/
CNTK
зеркало из
1
0
Форкнуть 0
Код Задачи Пакеты Проекты Релизы Вики Активность

Updated samples, remove bogus restriction from conv and pool nodes.

This commit is contained in:
Alexey Kamenev 2015-12-23 13:44:36 -08:00
Родитель 93ea2b519e
Коммит 9c745a516d
2 изменённых файлов: 5 добавлений и 64 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

63
Examples/Image/Miscellaneous/CIFAR-10/Macros.ndl
Просмотреть файл

@ -41,7 +41,7 @@ ResNetNode2(inp, outMap, inWCount, kW, kH, wScale, bValue, scValue)
isd1 = Parameter(outMap, 1, init = fixedValue, value = 0, needGradient = false)
c1 = Convolution(W1, inp, kW, kH, outMap, 1, 1, zeroPadding = true)
bn1 = BatchNormalization(c1, sc1, b1, m1, isd1, eval = false, spatial = true)
bn1 = BatchNormalization(c1, sc1, b1, m1, isd1, eval = false, spatial = true, expAvgFactor = 0.5)
y1 = RectifiedLinear(bn1);
W2 = Parameter(outMap, inWCount, init = Gaussian, initValueScale = wScale)
@ -51,64 +51,11 @@ ResNetNode2(inp, outMap, inWCount, kW, kH, wScale, bValue, scValue)
isd2 = Parameter(outMap, 1, init = fixedValue, value = 0, needGradient = false)
c2 = Convolution(W2, y1, kW, kH, outMap, 1, 1, zeroPadding = true)
bn2 = BatchNormalization(c2, sc2, b2, m2, isd2, eval = false, spatial = true)
bn2 = BatchNormalization(c2, sc2, b2, m2, isd2, eval = false, spatial = true, expAvgFactor = 0.5)
p = Plus(bn2, inp)
y2 = RectifiedLinear(p);
}
ResNetNode2Reduce(inp, outMap, inWCount, wCount, inDim, outDim, kW, kH, wScale, bValue, scValue)
{
W1 = Parameter(outMap, inWCount, init = Gaussian, initValueScale = wScale)
b1 = Parameter(outMap, 1, init = fixedValue, value = bValue)
sc1 = Parameter(outMap, 1, init = Gaussian, initValueScale = scValue)
m1 = Parameter(outMap, 1, init = fixedValue, value = 0, needGradient = false)
isd1 = Parameter(outMap, 1, init = fixedValue, value = 0, needGradient = false)
c1 = Convolution(W1, inp, kW, kH, outMap, 2, 2, zeroPadding = true)
bn1 = BatchNormalization(c1, sc1, b1, m1, isd1, eval = false, spatial = true)
y1 = RectifiedLinear(bn1);
W2 = Parameter(outMap, wCount, init = Gaussian, initValueScale = wScale)
b2 = Parameter(outMap, 1, init = fixedValue, value = bValue)
sc2 = Parameter(outMap, 1, init = Gaussian, initValueScale = scValue)
m2 = Parameter(outMap, 1, init = fixedValue, value = 0, needGradient = false)
isd2 = Parameter(outMap, 1, init = fixedValue, value = 0, needGradient = false)
c2 = Convolution(W2, y1, kW, kH, outMap, 1, 1, zeroPadding = true)
bn2 = BatchNormalization(c2, sc2, b2, m2, isd2, eval = false, spatial = true)
WP = Parameter(outDim, inDim)
t = Times(WP, inp, init = Gaussian, initValueScale = wScale)
p = Plus(bn2, t)
y2 = RectifiedLinear(p);
}
ResNetNode2MaxPool(inp, outMap, inWCount, wCount, kW, kH, wScale, bValue, scValue)
{
W1 = Parameter(outMap, inWCount, init = Gaussian, initValueScale = wScale)
b1 = Parameter(outMap, 1, init = fixedValue, value = bValue)
sc1 = Parameter(outMap, 1, init = Gaussian, initValueScale = scValue)
m1 = Parameter(outMap, 1, init = fixedValue, value = 0, needGradient = false)
isd1 = Parameter(outMap, 1, init = fixedValue, value = 0, needGradient = false)
c1 = Convolution(W1, inp, kW, kH, outMap, 2, 2, zeroPadding = true)
bn1 = BatchNormalization(c1, sc1, b1, m1, isd1, eval = false, spatial = true)
y1 = RectifiedLinear(bn1);
W2 = Parameter(outMap, wCount, init = Gaussian, initValueScale = wScale)
b2 = Parameter(outMap, 1, init = fixedValue, value = bValue)
sc2 = Parameter(outMap, 1, init = Gaussian, initValueScale = scValue)
m2 = Parameter(outMap, 1, init = fixedValue, value = 0, needGradient = false)
isd2 = Parameter(outMap, 1, init = fixedValue, value = 0, needGradient = false)
c2 = Convolution(W2, y1, kW, kH, outMap, 1, 1, zeroPadding = true)
bn2 = BatchNormalization(c2, sc2, b2, m2, isd2, eval = false, spatial = true)
imp = MaxPooling(inp, 2, 2, 2, 2)
imp2 = RowStack(imp, imp)
p = Plus(bn2, imp2)
y2 = RectifiedLinear(p);
}
ResNetNode2Conv(inp, outMap, inWCount, wCount, kW, kH, wScale, bValue, scValue, Wproj)
{
W1 = Parameter(outMap, inWCount, init = Gaussian, initValueScale = wScale)
@ -118,7 +65,7 @@ ResNetNode2Conv(inp, outMap, inWCount, wCount, kW, kH, wScale, bValue, scValue,
isd1 = Parameter(outMap, 1, init = fixedValue, value = 0, needGradient = false)
c1 = Convolution(W1, inp, kW, kH, outMap, 2, 2, zeroPadding = true)
bn1 = BatchNormalization(c1, sc1, b1, m1, isd1, eval = false, spatial = true)
bn1 = BatchNormalization(c1, sc1, b1, m1, isd1, eval = false, spatial = true, expAvgFactor = 0.5)
y1 = RectifiedLinear(bn1);
W2 = Parameter(outMap, wCount, init = Gaussian, initValueScale = wScale)
@ -128,9 +75,9 @@ ResNetNode2Conv(inp, outMap, inWCount, wCount, kW, kH, wScale, bValue, scValue,
isd2 = Parameter(outMap, 1, init = fixedValue, value = 0, needGradient = false)
c2 = Convolution(W2, y1, kW, kH, outMap, 1, 1, zeroPadding = true)
bn2 = BatchNormalization(c2, sc2, b2, m2, isd2, eval = false, spatial = true)
bn2 = BatchNormalization(c2, sc2, b2, m2, isd2, eval = false, spatial = true, expAvgFactor = 0.5)
cproj = Convolution(Wproj, inp, 2, 2, outMap, 2, 2, zeroPadding = false)
cproj = Convolution(Wproj, inp, 1, 1, outMap, 2, 2, zeroPadding = false)
p = Plus(bn2, cproj)
y2 = RectifiedLinear(p);
}

6
Source/ComputationNetworkLib/ConvolutionalNodes.h
Просмотреть файл

@ -171,9 +171,6 @@ namespace Microsoft { namespace MSR { namespace CNTK {
{
Base::Validate(isFinalValidationPass);
if (m_horizontalSubsample > m_kernelWidth || m_verticalSubsample > m_kernelHeight)
InvalidArgument("In ConvolutionNode horizontalSubsample must <= kernelWidth and verticalSubsample must <= kernelHeight.");
InferMBLayoutFromInputsForStandardCase();
InferImageDimsFromInputs();
@ -387,9 +384,6 @@ namespace Microsoft { namespace MSR { namespace CNTK {
{
Base::Validate(isFinalValidationPass);
if (m_horizontalSubsample > m_windowWidth || m_verticalSubsample > m_windowHeight)
InvalidArgument("PoolingNodeBase: horizontalSubsample must <= windowWidth and verticalSubsample must <= windowHeight.");
InferMBLayoutFromInputsForStandardCase();
InferImageDimsFromInputs();