/// <summary>
/// Dispose necessary objects.
/// </summary>
/// <param name="sender"></param>
/// <param name="e"></param>
private void Window_Closing(object sender, CancelEventArgs e)
{
camera.Stop();
camera.Dispose();
session.Dispose();
graph.Dispose();
}
public void Dispose()
{
Console.WriteLine("Finalizing TF model resources");
session.CloseSession();
session.Dispose();
Console.WriteLine("TF resources closed");
}
public void Dispose()
{
if (session != null)
{
session.Dispose();
}
if (graph != null)
{
graph.Dispose();
}
}
protected virtual void Dispose(bool disposing)
{
if (!_disposed)
{
if (disposing)
{
_session.Dispose();
_graph.Dispose();
}
_disposed = true;
}
}
public void TestCleanup()
{
if (s != null)
{
s.CloseSession();
s.Dispose();
s = null;
}
if (g != null)
{
g.Dispose();
g = null;
}
}
void IDisposable.Dispose()
{
if (_session != null)
{
_session.Dispose();
_session = null;
}
if (_graph != null)
{
_graph.Dispose();
_graph = null;
}
}
private void Dispose(bool disposing)
{
// Ensure that the Session is not null and it's handle is not Zero, as it may have already been disposed/finalized.
// Technically we shouldn't be calling this if disposing == false, since we're running in finalizer
// and the GC doesn't guarantee ordering of finalization of managed objects, but we have to make sure
// that the Session is closed before deleting our temporary directory.
try
{
if (Session?.Handle != IntPtr.Zero)
{
Session.CloseSession();
Session.Dispose();
}
}
finally
{
if (!string.IsNullOrEmpty(_savedModelPath) && _isTemporarySavedModel)
{
TensorFlowUtils.DeleteFolderWithRetries(_host, _savedModelPath);
}
}
}
public void Dispose()
{
_graph.Dispose();
_session.Dispose();
}
// run the CNN
void evaluate()
{
// only run CNN if we have enough accelerometer values
if (accelX.Count == inputWidth)
{
// convert from list to tensor
// if tensor is 1 under, add dummy last value
int i;
for (i = 0; i < accelX.Count; i++)
{
inputTensor[0, 0, i, 0] = accelX[i];
test = inputTensor[0, 0, i, 0];
}
if (i != inputWidth)
{
inputTensor[0, 0, inputWidth - 1, 0] = 0;
}
for (i = 0; i < accelY.Count; i++)
{
inputTensor[0, 0, i, 1] = accelY[i];
}
if (i != inputWidth)
{
inputTensor[0, 0, inputWidth - 1, 1] = 0;
}
for (i = 0; i < accelZ.Count; i++)
{
inputTensor[0, 0, i, 2] = accelZ[i];
}
if (i != inputWidth)
{
inputTensor[0, 0, inputWidth - 1, 2] = 0;
}
// tensor output variable
float[,] recurrentTensor;
// create tensorflow model
using (var graph = new TFGraph())
{
graph.Import(graphModel.bytes);
var session = new TFSession(graph);
var runner = session.GetRunner();
// do input tensor list to array and make it one dimensional
TFTensor input = inputTensor;
// set up input tensor and input
runner.AddInput(graph["input_placeholder_x"][0], input);
// set up output tensor
runner.Fetch(graph["output_node"][0]);
// run model
recurrentTensor = runner.Run()[0].GetValue() as float[, ];
here = true;
// dispose resources - keeps cnn from breaking down later
session.Dispose();
graph.Dispose();
}
// find the most confident answer
float highVal = 0;
int highInd = -1;
sum = 0f;
// *MAKE SURE ACTIVITYINDEXCHOICES MATCHES THE NUMBER OF CHOICES*
for (int j = 0; j < activityIndexChoices; j++)
{
confidence = recurrentTensor[0, j];
if (highInd > -1)
{
if (recurrentTensor[0, j] > highVal)
{
highVal = confidence;
highInd = j;
}
}
else
{
highVal = confidence;
highInd = j;
}
// debugging - sum should = 1 at the end
sum += confidence;
}
// debugging
test1 = recurrentTensor[0, 0];
test2 = recurrentTensor[0, 1];
test3 = recurrentTensor[0, 2];
// used in movement to see if we should be moving
index = highInd;
countCNN++;
}
}
public void Close()
{
session?.Dispose();
graph?.Dispose();
labels = null;
}
public TFTensor 转换方法()
{
//if (b)
//{
// contents = File.ReadAllBytes(file);
// // DecodeJpeg uses a scalar String-valued tensor as input.
// tensor = TFTensor.CreateString(contents);
// b = false;
//}
contents = File.ReadAllBytes(file);
//// DecodeJpeg uses a scalar String-valued tensor as input.
tensor = TFTensor.CreateString(contents);
//contents = null;
// Construct a graph to normalize the image 归一化
// Execute that graph to normalize this one image 执行图规范化这个形象
//using (var session = new TFSession(graph1))
//{
// var normalized = session.Run(
// inputs: new[] { input },
// inputValues: new[] { tensor },
// outputs: new[] { output });
// //tensor = null;
// return normalized[0];
//}
//using (var session = new TFSession(graph1))
//{
//var session = new TFSession(graph1);
// var runner = session.GetRunner();
// runner.AddInput(input, tensor);
// runner.Fetch(output);
// normalized = runner.Run();
// //session.CloseSession();
// session.DeleteSession();
// runner = null;
// //return normalized[0];
////}
//return 1;
using (var session = new TFSession(graph1))
{
try
{
var runner = session.GetRunner();
runner.AddInput(input, tensor);
runner.Fetch(output);
normalized = runner.Run();
return(normalized[0]);
}
catch (Exception e)
{
//Console.WriteLine(e);
throw;
}
finally
{
session.Dispose();
//session.CloseSession();
//session.DeleteSession();
}
}
//normalized = session.Run(
// inputs: new[] { input },
// inputValues: new[] { tensor },
// outputs: new[] { output });
//tensor = null;
//return normalized[0];
}
//double[] retResult;
public Int32 识别方法(out float ou)
{
tensor = _图片转换成张量类.转换方法();
var probabilities11 = ((float[][][][])tensor.GetValue(jagged: true))[0];
//tensor = CreateTensorFromImageFile(临时图片路径);
using (var session = new TFSession(graph))
{
try
{
var runner = session.GetRunner();
//runner.AddInput(graph["x_input"][0], tensor).Fetch(graph["softmax_linear/softmax_linear"][0]);
//runner.AddInput(graph["x_input"][0], tensor).Fetch(graph["softmax_z/softmax_z"][0]);
runner.AddInput(graph["x_input"][0], tensor);
runner.Fetch(graph["softmax_z/softmax_z"][0]);
output = runner.Run();
}
catch (Exception e)
{
Console.WriteLine(e);
throw;
}
finally
{
session.Dispose();
//session.CloseSession();
//session.DeleteSession();
}
}
//Random rd = new Random();
//ou = rd.Next(100, 500);
//return rd.Next(100, 500);
result = output[0];
rshape = result.Shape;
//var probabilities1 = ((float[][])result.GetValue(jagged: true))[0];
//var val1 = (float[,])result.GetValue(jagged: false);
//var val2 = (float[])result.GetValue(jagged: true);
if (result.NumDims != 2 || rshape[0] != 1)
{
var shape = "";
foreach (var d in rshape)
{
shape += $"{d} ";
}
shape = shape.Trim();
Console.WriteLine($"Error: expected to produce a [1 N] shaped tensor where N is the number of labels, instead it produced one with shape [{shape}]");
Environment.Exit(1);
}
jagged = true;
//jagged = false;
bestIdx = 0;
p = 0;
best = 0;
if (jagged)
{
var probabilities = ((float[][])result.GetValue(jagged: true))[0];
//double[] d = floatTodouble(probabilities);
//double[] retResult = Softmax(d);
for (int i = 0; i < probabilities.Length; i++)
{
if (probabilities[i] > best)
{
bestIdx = i;
//best = probabilities[i];
best = probabilities[i];
}
}
}
else
{
var val = (float[, ])result.GetValue(jagged: false);
for (int i = 0; i < val.GetLength(1); i++)
{
if (val[0, i] > best)
{
bestIdx = i;
best = val[0, i];
}
}
}
tensor = null;
result = null;
rshape = null;
ou = best;
return(bestIdx);
}
public void Dispose()
{
_graph?.Dispose();
_session?.Dispose();
}