///<summary>Выполняет прямой проход через кодирующую нейросеть.</summary>
///<param name="Content4_1">Контент со слоя ReLU4_1.</param>
///<param name="Style4_1">Стиль со слоя ReLU4_1.</param>
///<param name="Content5_1">Контент со слоя ReLU5_1.</param>
///<param name="Style5_1">Стиль со слоя ReLU5_1.</param>
public Tensor Stylize(Tensor Content4_1, Tensor Style4_1, Tensor Content5_1, Tensor Style5_1)
{
var Fc4_1 = Layers.Norm(Content4_1);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var Fs4_1 = Layers.Norm(Style4_1);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var Fc5_1 = Layers.Norm(Content5_1);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var Fs5_1 = Layers.Norm(Style5_1);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var f4_1 = Layers.Conv2D1x1(Fc4_1, this.Data.sanet4_1_f_Weights, this.Data.sanet4_1_f_Biases);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var g4_1 = Layers.Conv2D1x1(Fs4_1, this.Data.sanet4_1_g_Weights, this.Data.sanet4_1_g_Biases);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var h4_1 = Layers.Conv2D1x1(Style4_1, this.Data.sanet4_1_h_Weights, this.Data.sanet4_1_h_Biases);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var f5_1 = Layers.Conv2D1x1(Fc5_1, this.Data.sanet5_1_f_Weights, this.Data.sanet5_1_f_Biases);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var g5_1 = Layers.Conv2D1x1(Fs5_1, this.Data.sanet5_1_g_Weights, this.Data.sanet5_1_g_Biases);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var h5_1 = Layers.Conv2D1x1(Style5_1, this.Data.sanet5_1_h_Weights, this.Data.sanet5_1_h_Biases);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var S4_1 = Tensor.MatMul(f4_1.Flat().Transpose(), g4_1.Flat());
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
S4_1 = Layers.Softmax(S4_1);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var O4_1 = Layers.ElementwiseSum(Layers.Conv2D1x1(Tensor.MatMul(h4_1.Flat(), S4_1.Transpose()).Unflat(Content4_1.Width, Content4_1.Height), this.Data.sanet4_1_out_conv_Weights, this.Data.sanet4_1_out_conv_Biases), Content4_1);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var S5_1 = Tensor.MatMul(f5_1.Flat().Transpose(), g5_1.Flat());
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
S5_1 = Layers.Softmax(S5_1);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var O5_1 = Layers.ElementwiseSum(Layers.Conv2D1x1(Tensor.MatMul(h5_1.Flat(), S5_1.Transpose()).Unflat(Content5_1.Width, Content5_1.Height), this.Data.sanet5_1_out_conv_Weights, this.Data.sanet5_1_out_conv_Biases), Content5_1);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
var CS = Layers.Conv2D3x3(Layers.ElementwiseSum(O4_1, Layers.Upsample2D(O5_1)), this.Data.merge_conv_Weights, this.Data.merge_conv_Biases);
if (this.Step != null)
{
this.Step(1f / 17f * 100f);
}
return(CS);
}
///<summary>Выполняет прямой проход через декодирующую нейросеть.</summary>
///<param name="input">Входные данные.</param>
public Tensor Decode(Tensor input)
{
var Temp = input;
if (this.Depth == DecoderType.Conv5)
{
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv5_1_Weights, this.Data.Conv5_1_Biases);
if (this.Step != null)
{
this.Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(100f / 30f);
}
Temp = Layers.Upsample2D(Temp);
if (this.Step != null)
{
this.Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv4_4_Weights, this.Data.Conv4_4_Biases);
if (this.Step != null)
{
this.Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv4_3_Weights, this.Data.Conv4_3_Biases);
if (this.Step != null)
{
this.Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv4_2_Weights, this.Data.Conv4_2_Biases);
if (this.Step != null)
{
this.Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(100f / 30f);
}
}
if ((this.Depth == DecoderType.Conv4) || (this.Depth == DecoderType.Conv5))
{
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv4_1_Weights, this.Data.Conv4_1_Biases);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
}
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
}
}
Temp = Layers.Upsample2D(Temp);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
}
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_4_Weights, this.Data.Conv3_4_Biases);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
}
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
}
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_3_Weights, this.Data.Conv3_3_Biases);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
}
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
}
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_2_Weights, this.Data.Conv3_2_Biases);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
}
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
}
}
}
if ((this.Depth == DecoderType.Conv3) || (this.Depth == DecoderType.Conv4) || (this.Depth == DecoderType.Conv5))
{
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_1_Weights, this.Data.Conv3_1_Biases);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
case DecoderType.Conv3:
{
this.Step(100f / 12f);
break;
}
}
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
case DecoderType.Conv3:
{
this.Step(100f / 12f);
break;
}
}
}
Temp = Layers.Upsample2D(Temp);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
case DecoderType.Conv3:
{
this.Step(100f / 12f);
break;
}
}
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_2_Weights, this.Data.Conv2_2_Biases);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
case DecoderType.Conv3:
{
this.Step(100f / 12f);
break;
}
}
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
case DecoderType.Conv3:
{
this.Step(100f / 12f);
break;
}
}
}
}
if ((this.Depth == DecoderType.Conv2) || (this.Depth == DecoderType.Conv3) || (this.Depth == DecoderType.Conv4) || (this.Depth == DecoderType.Conv5))
{
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_1_Weights, this.Data.Conv2_1_Biases);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
case DecoderType.Conv3:
{
this.Step(100f / 12f);
break;
}
case DecoderType.Conv2:
{
this.Step(100f / 7f);
break;
}
}
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
case DecoderType.Conv3:
{
this.Step(100f / 12f);
break;
}
case DecoderType.Conv2:
{
this.Step(100f / 7f);
break;
}
}
}
Temp = Layers.Upsample2D(Temp);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
case DecoderType.Conv3:
{
this.Step(100f / 12f);
break;
}
case DecoderType.Conv2:
{
this.Step(100f / 7f);
break;
}
}
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_2_Weights, this.Data.Conv1_2_Biases);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
case DecoderType.Conv3:
{
this.Step(100f / 12f);
break;
}
case DecoderType.Conv2:
{
this.Step(100f / 7f);
break;
}
}
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
case DecoderType.Conv3:
{
this.Step(100f / 12f);
break;
}
case DecoderType.Conv2:
{
this.Step(100f / 7f);
break;
}
}
}
}
if ((this.Depth == DecoderType.Conv1) || (this.Depth == DecoderType.Conv2) || (this.Depth == DecoderType.Conv3) || (this.Depth == DecoderType.Conv4) || (this.Depth == DecoderType.Conv5))
{
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_1_Weights, this.Data.Conv1_1_Biases);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
case DecoderType.Conv3:
{
this.Step(100f / 12f);
break;
}
case DecoderType.Conv2:
{
this.Step(100f / 7f);
break;
}
case DecoderType.Conv1:
{
this.Step(100f / 2f);
break;
}
}
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
switch (this.Depth)
{
case DecoderType.Conv5:
{
this.Step(100f / 30f);
break;
}
case DecoderType.Conv4:
{
this.Step(100f / 21f);
break;
}
case DecoderType.Conv3:
{
this.Step(100f / 12f);
break;
}
case DecoderType.Conv2:
{
this.Step(100f / 7f);
break;
}
case DecoderType.Conv1:
{
this.Step(100f / 2f);
break;
}
}
}
}
return(Temp);
}
///<summary>Выполняет прямой проход через кодирующую нейросеть.</summary>
///<param name="input">Входные данные.</param>
public Tuple <Tensor, Tensor> Encode(Tensor input)
{
var Temp = input;
Temp = Layers.Conv2D1x1(input, this.Data.Conv0_Weights, this.Data.Conv0_Biases);
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_1_Weights, this.Data.Conv1_1_Biases);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_2_Weights, this.Data.Conv1_2_Biases);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.MaxPool2D(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_1_Weights, this.Data.Conv2_1_Biases);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_2_Weights, this.Data.Conv2_2_Biases);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.MaxPool2D(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_1_Weights, this.Data.Conv3_1_Biases);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_2_Weights, this.Data.Conv3_2_Biases);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_3_Weights, this.Data.Conv3_3_Biases);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_4_Weights, this.Data.Conv3_4_Biases);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.MaxPool2D(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv4_1_Weights, this.Data.Conv4_1_Biases);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
var Conv4_1 = Temp;
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv4_2_Weights, this.Data.Conv4_2_Biases);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv4_3_Weights, this.Data.Conv4_3_Biases);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv4_4_Weights, this.Data.Conv4_4_Biases);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.MaxPool2D(Temp);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv5_1_Weights, this.Data.Conv5_1_Biases);
if (Step != null)
{
Step(100f / 30f);
}
Temp = Layers.ReLU(Temp);
var Conv5_1 = Temp;
if (Step != null)
{
Step(100f / 30f);
}
return(new Tuple <Tensor, Tensor>(Conv4_1, Conv5_1));
}
///<summary>Выполняет прямой проход через кодирующую нейросеть.</summary>
///<param name="input">Входные данные.</param>
public Tensor Encode(Tensor input)
{
var Temp = input;
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_1_Weights, this.Data.Conv1_1_Biases);
if (this.Step != null)
{
this.Step(1f / 12f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 12f * 100f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_2_Weights, this.Data.Conv1_2_Biases);
if (this.Step != null)
{
this.Step(1f / 12f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 12f * 100f);
}
Temp = Layers.MaxPool2D(Temp);
if (this.Step != null)
{
this.Step(1f / 12f * 100f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_1_Weights, this.Data.Conv2_1_Biases);
if (this.Step != null)
{
this.Step(1f / 12f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 12f * 100f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_2_Weights, this.Data.Conv2_2_Biases);
if (this.Step != null)
{
this.Step(1f / 12f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 12f * 100f);
}
Temp = Layers.MaxPool2D(Temp);
if (this.Step != null)
{
this.Step(1f / 12f * 100f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_1_Weights, this.Data.Conv3_1_Biases);
if (this.Step != null)
{
this.Step(1f / 12f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 12f * 100f);
}
return(Temp);
}
///<summary>Выполняет прямой проход через декодирующую нейросеть.</summary>
///<param name="input">Входные данные.</param>
public Tensor Decode(Tensor Content, Tensor[] Styles)
{
var Temp = Layers.StyleDecorator(Content, Styles[3]);
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv4_1_Weights, this.Data.Conv4_1_Biases);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.Upsample2D(Temp);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_4_Weights, this.Data.Conv3_4_Biases);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_3_Weights, this.Data.Conv3_3_Biases);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_2_Weights, this.Data.Conv3_2_Biases);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.AdaIN(Temp, Styles[2]);
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_1_Weights, this.Data.Conv3_1_Biases);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.Upsample2D(Temp);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_2_Weights, this.Data.Conv2_2_Biases);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.AdaIN(Temp, Styles[1]);
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_1_Weights, this.Data.Conv2_1_Biases);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.Upsample2D(Temp);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_2_Weights, this.Data.Conv1_2_Biases);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.AdaIN(Temp, Styles[0]);
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_1_Weights, this.Data.Conv1_1_Biases);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
Temp = Layers.Conv2D7x7(Temp, this.Data.Conv_Out_Weights, this.Data.Conv_Out_Biases);
if (this.Step != null)
{
this.Step(1f / 22f * 100f);
}
return(Temp);
}
///<summary>Выполняет прямой проход через декодирующую нейросеть.</summary>
///<param name="input">Входные данные.</param>
public Tensor Decode(Tensor input)
{
var Temp = input;
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv4_1_Weights, this.Data.Conv4_1_Biases);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.Upsample2D(Temp);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_4_Weights, this.Data.Conv3_4_Biases);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_3_Weights, this.Data.Conv3_3_Biases);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_2_Weights, this.Data.Conv3_2_Biases);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_1_Weights, this.Data.Conv3_1_Biases);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.Upsample2D(Temp);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_2_Weights, this.Data.Conv2_2_Biases);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_1_Weights, this.Data.Conv2_1_Biases);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.Upsample2D(Temp);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_2_Weights, this.Data.Conv1_2_Biases);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(100f / 20f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_1_Weights, this.Data.Conv1_1_Biases);
if (this.Step != null)
{
this.Step(100f / 20f);
}
return(Temp);
}
///<summary>Выполняет прямой проход через кодирующую нейросеть.</summary>
///<param name="input">Входные данные.</param>
public Tensor Encode(Tensor input)
{
var Temp = input;
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_1_Weights, this.Data.Conv1_1_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_2_Weights, this.Data.Conv1_2_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.AvgPool2D(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_1_Weights, this.Data.Conv2_1_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_2_Weights, this.Data.Conv2_2_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.AvgPool2D(Temp);
if (this.Step != null)
{
//this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_1_Weights, this.Data.Conv3_1_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_2_Weights, this.Data.Conv3_2_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_3_Weights, this.Data.Conv3_3_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_4_Weights, this.Data.Conv3_4_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.AvgPool2D(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv4_1_Weights, this.Data.Conv4_1_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
return(Temp);
}
///<summary>Выполняет прямой проход через кодирующую нейросеть.</summary>
///<param name="input">Входные данные.</param>
public Tensor[] EncodeStyle(Tensor input)
{
var Temp = input;
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_1_Weights, this.Data.Conv1_1_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
var Conv1 = Temp;
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv1_2_Weights, this.Data.Conv1_2_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.AvgPool2D(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_1_Weights, this.Data.Conv2_1_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
var Conv2 = Temp;
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv2_2_Weights, this.Data.Conv2_2_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.AvgPool2D(Temp);
if (this.Step != null)
{
//this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_1_Weights, this.Data.Conv3_1_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
var Conv3 = Temp;
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_2_Weights, this.Data.Conv3_2_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_3_Weights, this.Data.Conv3_3_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv3_4_Weights, this.Data.Conv3_4_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.AvgPool2D(Temp);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.Conv2D3x3(Temp, this.Data.Conv4_1_Weights, this.Data.Conv4_1_Biases);
if (this.Step != null)
{
this.Step(5f);
}
Temp = Layers.ReLU(Temp);
var Conv4 = Temp;
if (this.Step != null)
{
this.Step(5f);
}
return(new Tensor[4] {
Conv1, Conv2, Conv3, Conv4
});
}
