public async Task LoadMtcnnRNetNetworkAsync()
{
using (var stream = this.GetType().Assembly.GetManifestResourceStream("TensorFlowCore.Test.mtcnn2.json"))
using (var textReader = new StreamReader(stream))
{
string json = await textReader.ReadToEndAsync().ConfigureAwait(false);
var obj = JsonConvert.DeserializeObject <LayerSerializationModel[]>(json);
Network network = new Network();
var conv1 = obj.Single(o => o.Name == "conv1");
var conv2 = obj.Single(o => o.Name == "conv2");
var conv3 = obj.Single(o => o.Name == "conv3");
var conv41 = obj.Single(o => o.Name == "conv4-1");
var conv42 = obj.Single(o => o.Name == "conv4-2");
var prelu1 = obj.Single(o => o.Name == "PReLU1");
var prelu2 = obj.Single(o => o.Name == "PReLU2");
var prelu3 = obj.Single(o => o.Name == "PReLU3");
network.AddInput("data", 29, 29, 3, 1)
.ContinueWithConv2D("conv1", (3, 3, 10), (1, 1), PaddingType.Same, conv1.Weights, biases: conv1.Biases, addReLu: false)
.ContinueWithPRelu("PReLU1", prelu1.Alpha)
.ContinueWithConv2D("conv2", (3, 3, 16), (1, 1), PaddingType.Same, conv2.Weights, biases: conv2.Biases, addReLu: false)
.ContinueWithPRelu("PReLU2", prelu2.Alpha)
.ContinueWithConv2D("conv3", (3, 3, 32), (1, 1), PaddingType.Same, conv3.Weights, biases: conv3.Biases, addReLu: false)
.ContinueWithPRelu("PReLU3", prelu3.Alpha)
.ContinueWithConv2D("conv4-1", (1, 1, 2), (1, 1), PaddingType.Same, conv41.Weights, biases: conv41.Biases, addReLu: false)
.ContinueWithMultidimensionalSoftMax("prob1", 3);
network.FromLayer("PReLU3")
.ContinueWithConv2D("conv4-2", (1, 1, 4), (1, 1), PaddingType.Same, conv42.Weights, biases: conv42.Biases, addReLu: false);
}
}
public async Task GivenIHaveAnMtcnnPNetDefinition_WhenIRunItWithValues_ThenWeObtainTheExpectedValues()
{
using (var stream = File.OpenRead("TensorflowSharpCore/Resources/mtcnn1.json"))
using (var textReader = new StreamReader(stream))
{
string json = await textReader.ReadToEndAsync().ConfigureAwait(false);
var obj = JsonConvert.DeserializeObject <LayerSerializationModel[]>(json);
var network = new Network();
var conv1 = obj.Single(o => o.Name == "conv1");
var conv2 = obj.Single(o => o.Name == "conv2");
var conv3 = obj.Single(o => o.Name == "conv3");
var conv41 = obj.Single(o => o.Name == "conv4-1");
var conv42 = obj.Single(o => o.Name == "conv4-2");
var prelu1 = obj.Single(o => o.Name == "PReLU1");
var prelu2 = obj.Single(o => o.Name == "PReLU2");
var prelu3 = obj.Single(o => o.Name == "PReLU3");
network.AddInput("data", TfConstants.NullDimension, TfConstants.NullDimension, 3, 1)
.ContinueWithConv2D("conv1", (3, 3, 10), (1, 1), PaddingType.Valid, conv1.Weights, biases: conv1.Biases, addReLu: false)
.ContinueWithPRelu("PReLU1", prelu1.Alpha)
.ContinueWithConv2D("conv2", (3, 3, 16), (1, 1), PaddingType.Same, conv2.Weights, biases: conv2.Biases, addReLu: false)
.ContinueWithPRelu("PReLU2", prelu2.Alpha)
.ContinueWithConv2D("conv3", (3, 3, 32), (1, 1), PaddingType.Same, conv3.Weights, biases: conv3.Biases, addReLu: false)
.ContinueWithPRelu("PReLU3", prelu3.Alpha)
.ContinueWithConv2D("conv4-1", (1, 1, 2), (1, 1), PaddingType.Same, conv41.Weights, biases: conv41.Biases)
.ContinueWithMultidimensionalSoftMax("prob1", 3);
network.FromLayer("PReLU3")
.ContinueWithConv2D("conv4-2", (1, 1, 4), (1, 1), PaddingType.Same, conv42.Weights, biases: conv42.Biases);
float[,,,] inputValues;
//using (var imageResource = new TestResource(GetType().Assembly, "FaceRecognition.UnitTests.FaceDetections.Algorithms.Resources.image_020_1.jpg"))
//using (var imageStream = File.OpenRead(imageResource.ResourcePath))
//{
// var image = Image.Load(imageStream);
// var baseImage = await BaseImage.CreateAsync(image, "image_020_1.jpg").ConfigureAwait(false);
// using (var inputArrayValues = baseImage.ToColorPixelsArray(615, 384))
// {
// inputValues = inputArrayValues.NormalizeBatch();
// }
//}
using (var jsonResurce = new TestResource(GetType().Assembly, "FaceRecognition.UnitTests.TensorflowSharpCore.Resources.data.json"))
{
inputValues = JsonConvert.DeserializeObject <float[, , , ]>(File.ReadAllText(jsonResurce.ResourcePath));
}
var networkResults = network.Run(new { data = inputValues }, new[] { "conv4-2", "prob1" });
var points = new List <Vector2>();
var prob = networkResults.ElementAt(1);
System.Diagnostics.Debug.WriteLine($"{points.Count} boxes found");
networkResults.Count().Should().Be(2);
}
}
