public override void Backward(SuperArray outputgrad)
{
int pad = 0;
if (Padding == PaddingType.Same)
{
pad = 1;
}
else if (Padding == PaddingType.Full)
{
pad = 2;
}
var dout_flat = outputgrad.Transpose(2, 0, 1).Reshape(Filters, -1);
var dW = Ops.Dot(dout_flat, xCols.Transpose());
dW = dW.Reshape(Params["w"].Data.Shape);
var db = Ops.Sum(outputgrad, new uint[] { 0, 1, 2 }).Reshape(Filters, -1);
var W_flat = Params["w"].Data.Reshape(Filters, -1);
var dX_col = Ops.Dot(W_flat.Transpose(), dout_flat);
Input.Grad = ImUtil.Col2Im(dX_col, Input.Data.Shape.Dims, Tuple.Create(KernalSize, KernalSize), pad, Strides);
Params["w"].Grad = dW;
if (UseBias)
{
Params["b"].Grad = db;
}
}
public override void Forward(SuperArray x)
{
base.Forward(x);
var(n, c, h, w) = x.GetConv2DShape();
int pad = 0;
if (Padding == PaddingType.Same)
{
pad = 1;
}
else if (Padding == PaddingType.Full)
{
pad = 2;
}
var h_out = (h - PoolSize.Item1 + 2 * pad) / Strides + 1;
var w_out = (w - PoolSize.Item2 + 2 * pad) / Strides + 1;
var x_reshaped = x.Reshape(n * c, 1, h, w);
xCols = ImUtil.Im2Col(x_reshaped, PoolSize, pad, Strides);
Output = Ops.Max(xCols, 0);
Output = Output.Reshape(h_out, w_out, n, c).Transpose(2, 3, 0, 1);
}
public override void Forward(SuperArray x)
{
base.Forward(x);
var(n, c, h, w) = x.GetConv2DShape();
Parameter weight = BuildParam("w", new Shape(Filters, c, KernalSize.Item1, KernalSize.Item2), KernalInitializer, KernalConstraint, KernalRegularizer);
Parameter bias = null;
if (UseBias)
{
bias = BuildParam("b", new Shape(Filters, 1), BiasInitializer, BiasConstraint, BiasRegularizer);
}
int pad = 0;
if (Padding == PaddingType.Same)
{
pad = 1;
}
else if (Padding == PaddingType.Full)
{
pad = 2;
}
var dialatedKernel = Tuple.Create(((KernalSize.Item1 - 1) * DialationRate.Item1 + 1), ((KernalSize.Item2 - 1) * DialationRate.Item2 + 1));
var h_out = (h - dialatedKernel.Item1 + 2 * pad) / Strides + 1;
var w_out = (w - dialatedKernel.Item2 + 2 * pad) / Strides + 1;
var wRows = weight.Data.Reshape(Filters, -1);
xCols = ImUtil.Im2Col(x, dialatedKernel, pad, Strides);
Output = Ops.Dot(wRows, xCols);
if (UseBias)
{
Output = Output + bias.Data;
}
Output = Output.Reshape(Filters, h_out, w_out, n);
Output = Output.Transpose(3, 0, 1, 2);
}
public override void Backward(SuperArray outputgrad)
{
SuperArray dX_col = Ops.Constant(0, xCols.Shape);
var(n, c, h, w) = Input.Data.GetConv2DShape();
int pad = 0;
if (Padding == PaddingType.Same)
{
pad = 1;
}
else if (Padding == PaddingType.Full)
{
pad = 2;
}
var dout_flat = outputgrad.Transpose(2, 3, 0, 1);
var dX = ImUtil.Col2Im(dout_flat, Input.Data.Shape.Dims, PoolSize, pad, Strides);
Input.Grad = dX.Reshape(n, c, h, w);
}
public override void Forward(SuperArray x)
{
base.Forward(x);
var(n, c, s) = x.GetConv1DShape();
Parameter weight = BuildParam("w", new Shape(Filters, c, KernalSize), KernalInitializer, KernalConstraint, KernalRegularizer);
Parameter bias = null;
if (UseBias)
{
bias = BuildParam("b", new Shape(Filters, 1), BiasInitializer, BiasConstraint, BiasRegularizer);
}
int pad = 0;
if (Padding == PaddingType.Same)
{
pad = 1;
}
else if (Padding == PaddingType.Full)
{
pad = 2;
}
KernalSize = (KernalSize - 1) * DilationRate + 1;
var steps_out = (s - KernalSize + 2 * pad) / Strides + 1;
xCols = ImUtil.Im2Col(x, Tuple.Create(KernalSize, KernalSize), pad, Strides);
var wRows = weight.Data.Reshape(Filters, -1);
Output = Ops.Dot(wRows, xCols);
if (UseBias)
{
Output = Output + bias.Data;
}
Output = Output.Reshape(Filters, steps_out, n);
Output = Output.Transpose(2, 0, 1);
}
public override void Forward(SuperArray x)
{
base.Forward(x);
var(n, c, s) = x.GetConv1DShape();
int pad = 0;
if (Padding == PaddingType.Same)
{
pad = 1;
}
else if (Padding == PaddingType.Full)
{
pad = 2;
}
var s_out = (s - PoolSize) / Strides + 1;
var x_reshaped = x.Reshape(n * c, 1, s);
xCols = ImUtil.Im2Col(x_reshaped, Tuple.Create(PoolSize, PoolSize), pad, Strides);
Output = Ops.Mean(xCols, 0);
Output = Output.Reshape(s_out, n, c).Transpose(2, 0, 1);
}
