// https://github.com/keras-team/keras/blob/49f5b931410bc2e56378f20a15e8ac919e0efb88/keras/applications/vgg16.py
// WEIGHTS_PATH_NO_TOP = 'https://github.com/fchollet/deep-learning-models/releases/download/v0.1/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5'
/// <summary>
/// モデル作って返します。
/// </summary>
/// <returns></returns>
public static Model CreateModel(string weightFile)
{
var img_input = Python.Input(new int?[] { 224, 224, 3 });
var x = new Conv2D(64, new[] { 3, 3 }, activation: new ReLU(), padding: PaddingType.Same).Call(img_input);
x = new Conv2D(64, new[] { 3, 3 }, activation: new ReLU(), padding: PaddingType.Same).Call(x);
x = new MaxPooling2D(new int[] { 2, 2 }).Call(x);
x = new Conv2D(128, new[] { 3, 3 }, activation: new ReLU(), padding: PaddingType.Same).Call(x);
x = new Conv2D(128, new[] { 3, 3 }, activation: new ReLU(), padding: PaddingType.Same).Call(x);
x = new MaxPooling2D(new int[] { 2, 2 }).Call(x);
x = new Conv2D(256, new[] { 3, 3 }, activation: new ReLU(), padding: PaddingType.Same).Call(x);
x = new Conv2D(256, new[] { 3, 3 }, activation: new ReLU(), padding: PaddingType.Same).Call(x);
x = new Conv2D(256, new[] { 3, 3 }, activation: new ReLU(), padding: PaddingType.Same).Call(x);
x = new MaxPooling2D(new int[] { 2, 2 }).Call(x);
x = new Conv2D(512, new[] { 3, 3 }, activation: new ReLU(), padding: PaddingType.Same).Call(x);
x = new Conv2D(512, new[] { 3, 3 }, activation: new ReLU(), padding: PaddingType.Same).Call(x);
x = new Conv2D(512, new[] { 3, 3 }, activation: new ReLU(), padding: PaddingType.Same).Call(x);
x = new MaxPooling2D(new int[] { 2, 2 }).Call(x);
x = new Conv2D(512, new[] { 3, 3 }, activation: new ReLU(), padding: PaddingType.Same).Call(x);
x = new Conv2D(512, new[] { 3, 3 }, activation: new ReLU(), padding: PaddingType.Same).Call(x);
x = new Conv2D(512, new[] { 3, 3 }, activation: new ReLU(), padding: PaddingType.Same).Call(x);
x = new MaxPooling2D(new int[] { 2, 2 }).Call(x);
var model = new Model(img_input, x, "vgg16");
model.LoadWeightsH5(weightFile);
return(model);
}
/// <summary>
/// Max pooling layer for 2D inputs (e.g. images).
/// </summary>
/// <param name="inputs">The tensor over which to pool. Must have rank 4.</param>
/// <param name="pool_size"></param>
/// <param name="strides"></param>
/// <param name="padding"></param>
/// <param name="data_format"></param>
/// <param name="name"></param>
/// <returns></returns>
public Tensor max_pooling2d(Tensor inputs,
int[] pool_size,
int[] strides,
string padding = "valid",
string data_format = "channels_last",
string name = null)
{
var layer = new MaxPooling2D(pool_size: pool_size,
strides: strides,
padding: padding,
data_format: data_format,
name: name);
return(layer.apply(inputs).Item1);
}
public ILayer CreateProduct(IKernelDescriptor descriptor)
{
if (descriptor is MaxPooling2D)
{
MaxPooling2D pool = descriptor as MaxPooling2D;
ILayer layer = new MaxPool2DLayer(pool.PaddingVertical, pool.PaddingHorizontal,
pool.StrideVertical, pool.StrideHorizontal,
pool.KernelHeight, pool.KernelWidth);
return(layer);
}
return(null);
}
/// <summary>
/// Max pooling layer for 2D inputs (e.g. images).
/// </summary>
/// <param name="inputs">The tensor over which to pool. Must have rank 4.</param>
/// <param name="pool_size"></param>
/// <param name="strides"></param>
/// <param name="padding"></param>
/// <param name="data_format"></param>
/// <param name="name"></param>
/// <returns></returns>
public Tensor max_pooling2d(Tensor inputs,
int[] pool_size,
int[] strides,
string padding = "valid",
string data_format = "channels_last",
string name = null)
{
var layer = new MaxPooling2D(new MaxPooling2DArgs
{
PoolSize = pool_size,
Strides = strides,
Padding = padding,
DataFormat = data_format,
Name = name
});
return(layer.Apply(inputs));
}
public static void Run()
{
Stopwatch sw = new Stopwatch();
NdArray inputArrayCpu = new NdArray(Initializer.GetRealArray(INPUT_SIZE));
NdArray inputArrayGpu = new NdArray(Initializer.GetRealArray(INPUT_SIZE));
//Linear
Linear linear = new Linear(INPUT_SIZE, OUTPUT_SIZE);
Console.WriteLine("◆" + linear.Name);
sw.Restart();
NdArray[] gradArrayCpu = linear.Forward(inputArrayCpu);
sw.Stop();
Console.WriteLine("Forward [Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradArrayCpu[0].Grad = gradArrayCpu[0].Data; //DataをGradとして使用
sw.Restart();
linear.Backward(gradArrayCpu);
sw.Stop();
Console.WriteLine("Backward[Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
if (linear.SetGpuEnable(true))
{
sw.Restart();
NdArray[] gradArrayGpu = linear.Forward(inputArrayGpu);
sw.Stop();
Console.WriteLine("Forward [Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradArrayGpu[0].Grad = gradArrayGpu[0].Data;
sw.Restart();
linear.Backward(gradArrayGpu);
sw.Stop();
Console.WriteLine("Backward[Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
}
//Tanh
TanhActivation tanh = new TanhActivation();
Console.WriteLine("\n◆" + tanh.Name);
sw.Restart();
gradArrayCpu = tanh.Forward(inputArrayCpu);
sw.Stop();
Console.WriteLine("Forward [Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradArrayCpu[0].Grad = gradArrayCpu[0].Data;
sw.Restart();
tanh.Backward(gradArrayCpu);
sw.Stop();
Console.WriteLine("Backward[Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
if (tanh.SetGpuEnable(true))
{
sw.Restart();
NdArray[] gradArrayGpu = tanh.Forward(inputArrayGpu);
sw.Stop();
Console.WriteLine("Forward [Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradArrayGpu[0].Grad = gradArrayGpu[0].Data;
sw.Restart();
tanh.Backward(gradArrayGpu);
sw.Stop();
Console.WriteLine("Backward[Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
}
//Sigmoid
Sigmoid sigmoid = new Sigmoid();
Console.WriteLine("\n◆" + sigmoid.Name);
sw.Restart();
gradArrayCpu = sigmoid.Forward(inputArrayCpu);
sw.Stop();
Console.WriteLine("Forward [Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradArrayCpu[0].Grad = gradArrayCpu[0].Data;
sw.Restart();
sigmoid.Backward(gradArrayCpu);
sw.Stop();
Console.WriteLine("Backward[Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
if (sigmoid.SetGpuEnable(true))
{
sw.Restart();
NdArray[] gradArrayGpu = sigmoid.Forward(inputArrayGpu);
sw.Stop();
Console.WriteLine("Forward [Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradArrayGpu[0].Grad = gradArrayGpu[0].Data;
sw.Restart();
sigmoid.Backward(gradArrayGpu);
sw.Stop();
Console.WriteLine("Backward[Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
}
//ReLU
ReLU relu = new ReLU();
Console.WriteLine("\n◆" + relu.Name);
sw.Restart();
gradArrayCpu = relu.Forward(inputArrayCpu);
sw.Stop();
Console.WriteLine("Forward [Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradArrayCpu[0].Grad = gradArrayCpu[0].Data;
sw.Restart();
relu.Backward(gradArrayCpu);
sw.Stop();
Console.WriteLine("Backward[Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
if (relu.SetGpuEnable(true))
{
sw.Restart();
NdArray[] gradArrayGpu = relu.Forward(inputArrayGpu);
sw.Stop();
Console.WriteLine("Forward [Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradArrayGpu[0].Grad = gradArrayGpu[0].Data;
sw.Restart();
relu.Backward(gradArrayGpu);
sw.Stop();
Console.WriteLine("Backward[Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
}
//LeakyReLU
LeakyReLU leakyRelu = new LeakyReLU();
Console.WriteLine("\n◆" + leakyRelu.Name);
sw.Restart();
gradArrayCpu = leakyRelu.Forward(inputArrayCpu);
sw.Stop();
Console.WriteLine("Forward [Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradArrayCpu[0].Grad = gradArrayCpu[0].Data;
sw.Restart();
leakyRelu.Backward(gradArrayCpu);
sw.Stop();
Console.WriteLine("Backward[Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
if (leakyRelu.SetGpuEnable(true))
{
sw.Restart();
NdArray[] gradArrayGpu = leakyRelu.Forward(inputArrayGpu);
sw.Stop();
Console.WriteLine("Forward [Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradArrayGpu[0].Grad = gradArrayGpu[0].Data;
sw.Restart();
leakyRelu.Backward(gradArrayGpu);
sw.Stop();
Console.WriteLine("Backward[Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
}
NdArray inputImageArrayGpu = new NdArray(Initializer.GetRealArray(3 * 256 * 256 * 5), new[] { 3, 256, 256 }, 5);
NdArray inputImageArrayCpu = new NdArray(Initializer.GetRealArray(3 * 256 * 256 * 5), new[] { 3, 256, 256 }, 5);
//MaxPooling
MaxPooling2D maxPooling2D = new MaxPooling2D(3);
Console.WriteLine("\n◆" + maxPooling2D.Name);
sw.Restart();
NdArray[] gradImageArrayCpu = maxPooling2D.Forward(inputImageArrayCpu);
sw.Stop();
Console.WriteLine("Forward [Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradImageArrayCpu[0].Grad = gradImageArrayCpu[0].Data;
sw.Restart();
maxPooling2D.Backward(gradImageArrayCpu);
sw.Stop();
Console.WriteLine("Backward[Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
if (maxPooling2D.SetGpuEnable(true))
{
sw.Restart();
maxPooling2D.Forward(inputImageArrayGpu);
sw.Stop();
Console.WriteLine("Forward [Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
//メモリ転送のみのため実装がない
Console.WriteLine("Backward[Gpu] : None");
}
//Conv2D
Convolution2D conv2d = new Convolution2D(3, 3, 3);
Console.WriteLine("\n◆" + conv2d.Name);
sw.Restart();
gradImageArrayCpu = conv2d.Forward(inputImageArrayCpu);
sw.Stop();
Console.WriteLine("Forward [Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradImageArrayCpu[0].Grad = gradImageArrayCpu[0].Data;
sw.Restart();
conv2d.Backward(gradImageArrayCpu);
sw.Stop();
Console.WriteLine("Backward[Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
if (conv2d.SetGpuEnable(true))
{
sw.Restart();
NdArray[] gradImageArrayGpu = conv2d.Forward(inputImageArrayGpu);
sw.Stop();
Console.WriteLine("Forward [Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradImageArrayGpu[0].Grad = gradImageArrayGpu[0].Data;
sw.Restart();
conv2d.Backward(gradImageArrayGpu);
sw.Stop();
Console.WriteLine("Backward[Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
}
//Deconv2D
Deconvolution2D deconv2d = new Deconvolution2D(3, 3, 3);
Console.WriteLine("\n◆" + deconv2d.Name);
sw.Restart();
gradImageArrayCpu = deconv2d.Forward(inputImageArrayCpu);
sw.Stop();
Console.WriteLine("Forward [Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradImageArrayCpu[0].Grad = gradImageArrayCpu[0].Data;
sw.Restart();
deconv2d.Backward(gradImageArrayCpu);
sw.Stop();
Console.WriteLine("Backward[Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
if (deconv2d.SetGpuEnable(true))
{
sw.Restart();
NdArray[] gradImageArrayGpu = deconv2d.Forward(inputImageArrayGpu);
sw.Stop();
Console.WriteLine("Forward [Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradImageArrayGpu[0].Grad = gradImageArrayGpu[0].Data;
sw.Restart();
deconv2d.Backward(gradImageArrayGpu);
sw.Stop();
Console.WriteLine("Backward[Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
}
//Dropout
Dropout dropout = new Dropout();
Console.WriteLine("\n◆" + dropout.Name);
sw.Restart();
gradArrayCpu = dropout.Forward(inputArrayCpu);
sw.Stop();
Console.WriteLine("Forward [Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradArrayCpu[0].Grad = gradArrayCpu[0].Data;
sw.Restart();
dropout.Backward(gradArrayCpu);
sw.Stop();
Console.WriteLine("Backward[Cpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
if (dropout.SetGpuEnable(true))
{
sw.Restart();
NdArray[] gradArrayGpu = dropout.Forward(inputArrayGpu);
sw.Stop();
Console.WriteLine("Forward [Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
gradArrayGpu[0].Grad = gradArrayGpu[0].Data;
sw.Restart();
dropout.Backward(gradArrayGpu);
sw.Stop();
Console.WriteLine("Backward[Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
}
}
private List <IKernelDescriptor> ReadDescriptors(JObject model)
{
List <IKernelDescriptor> dscps = model.SelectToken("descriptors").Select(layer => {
IKernelDescriptor descriptor = null;
String layerName = (String)layer.SelectToken("layer");
switch (layerName)
{
case "AvgPooling1D":
descriptor = new AvgPooling1D(
(int)layer.SelectToken("padding"),
(int)layer.SelectToken("stride"),
(int)layer.SelectToken("kernel_size"));
break;
case "GlobalAveragePooling1D":
descriptor = new GlobalAvgPooling1D();
break;
case "AvgPooling2D":
descriptor = new AvgPooling2D((int)layer.SelectToken("padding_vl"), (int)layer.SelectToken("padding_hz"),
(int)layer.SelectToken("stride_vl"), (int)layer.SelectToken("stride_hz"),
(int)layer.SelectToken("kernel_height"), (int)layer.SelectToken("kernel_width"));
break;
case "GlobalAveragePooling2D":
descriptor = new GlobalAvgPooling2D();
break;
case "BatchNormalization":
descriptor = new BatchNormalization(
(int)layer.SelectToken("epsilon"));
break;
case "Cropping1D":
descriptor = new Cropping1D(
(int)layer.SelectToken("trimBegin"),
(int)layer.SelectToken("trimEnd"));
break;
case "Cropping2D":
descriptor = new Cropping2D(
(int)layer.SelectToken("topTrim"),
(int)layer.SelectToken("bottomTrim"),
(int)layer.SelectToken("leftTrim"),
(int)layer.SelectToken("rightTrim"));
break;
case "MaxPooling1D":
descriptor = new MaxPooling1D(
(int)layer.SelectToken("padding"),
(int)layer.SelectToken("stride"),
(int)layer.SelectToken("kernel_size"));
break;
case "GlobalMaxPooling1D":
descriptor = new GlobalMaxPooling1D();
break;
case "MaxPooling2D":
descriptor = new MaxPooling2D((int)layer.SelectToken("padding_vl"), (int)layer.SelectToken("padding_hz"),
(int)layer.SelectToken("stride_vl"), (int)layer.SelectToken("stride_hz"),
(int)layer.SelectToken("kernel_height"), (int)layer.SelectToken("kernel_width"));
break;
case "GlobalMaxPooling2D":
descriptor = new GlobalMaxPooling2D();
break;
case "Convolution1D":
descriptor = new Convolution1D(
(int)layer.SelectToken("padding"),
(int)layer.SelectToken("stride"),
(int)layer.SelectToken("kernel_size"),
(int)layer.SelectToken("kernel_num"));
break;
case "Convolution2D":
descriptor = new Convolution2D((int)layer.SelectToken("padding_vl"), (int)layer.SelectToken("padding_hz"),
(int)layer.SelectToken("stride_vl"), (int)layer.SelectToken("stride_hz"),
(int)layer.SelectToken("kernel_height"), (int)layer.SelectToken("kernel_width"),
(int)layer.SelectToken("kernel_num"));
break;
case "Dense2D":
descriptor = new Dense2D((int)layer.SelectToken("units"));
break;
case "Input2D":
descriptor = new Input2D((int)layer.SelectToken("height"), (int)layer.SelectToken("width"),
(int)layer.SelectToken("channel"), (int)layer.SelectToken("batch"));
break;
case "Bias2D":
descriptor = new Bias2D();
break;
case "Permute":
descriptor = new Permute(
(int)layer.SelectToken("dim1"),
(int)layer.SelectToken("dim2"),
(int)layer.SelectToken("dim3"));
break;
case "Reshape":
descriptor = new Reshape2D(
(int)layer.SelectToken("height"),
(int)layer.SelectToken("width"),
(int)layer.SelectToken("channel"),
1);
break;
case "RepeatVector":
descriptor = new RepeatVector(
(int)layer.SelectToken("num"));
break;
case "SimpleRNN":
descriptor = new SimpleRNN(
(int)layer.SelectToken("units"),
(int)layer.SelectToken("input_dim"),
ANR((string)layer.SelectToken("activation")));
break;
case "LSTM":
descriptor = new LSTM(
(int)layer.SelectToken("units"),
(int)layer.SelectToken("input_dim"),
ANR((string)layer.SelectToken("activation")),
ANR((string)layer.SelectToken("rec_act")));
break;
case "GRU":
descriptor = new GRU(
(int)layer.SelectToken("units"),
(int)layer.SelectToken("input_dim"),
ANR((string)layer.SelectToken("activation")),
ANR((string)layer.SelectToken("rec_act")));
break;
case "ELu":
descriptor = new ELu(1);
break;
case "HardSigmoid":
descriptor = new HardSigmoid();
break;
case "ReLu":
descriptor = new ReLu();
break;
case "Sigmoid":
descriptor = new Sigmoid();
break;
case "Flatten":
descriptor = new Flatten();
break;
case "Softmax":
descriptor = new Softmax();
break;
case "SoftPlus":
descriptor = new SoftPlus();
break;
case "SoftSign":
descriptor = new Softsign();
break;
case "TanH":
descriptor = new TanH();
break;
default:
throw new Exception("Unknown layer type!");
}
return(descriptor);
}).ToList();
return(dscps);
}