private async Task LoadModelAsync()
{
if (this.model != null)
{
return;
}
await Dispatcher.RunAsync(CoreDispatcherPriority.Normal, () => StatusBlock.Text = $"Loading { MODEL_FILENAME } ... patience ");
try
{
// Load Model
var modelFile = await StorageFile.GetFileFromApplicationUriAsync(new Uri($"ms-appx:///Assets/{ MODEL_FILENAME }"));
this.model = await LearningModelPreview.LoadModelFromStorageFileAsync(modelFile);
// Retrieve model input and output variable descriptions (we already know the model takes an image in and outputs a tensor)
var inputFeatures = this.model.Description.InputFeatures.ToList();
var outputFeatures = this.model.Description.OutputFeatures.ToList();
this.inputImageDescription =
inputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image)
as ImageVariableDescriptorPreview;
this.outputTensorDescription =
outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Tensor)
as TensorVariableDescriptorPreview;
}
catch (Exception ex)
{
await Dispatcher.RunAsync(CoreDispatcherPriority.Normal, () => StatusBlock.Text = $"error: {ex.Message}");
model = null;
}
}
public async Task LoadModelAsync()
{
try
{
// Load Model
var modelFile = await StorageFile.GetFileFromApplicationUriAsync(
new Uri($"ms-appx:///Assets/FERPlus.onnx"));
model = await LearningModelPreview.LoadModelFromStorageFileAsync(modelFile);
// Retrieve model input and output variable descriptions (we already know
// the model takes an image in and outputs a tensor)
var inputFeatures = model.Description.InputFeatures.ToList();
var outputFeatures = model.Description.OutputFeatures.ToList();
inputImageDescription =
inputFeatures.FirstOrDefault(
feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image)
as ImageVariableDescriptorPreview;
outputTensorDescription =
outputFeatures.FirstOrDefault(
feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Tensor)
as TensorVariableDescriptorPreview;
}
catch (Exception)
{
model = null;
throw;
}
}
public async void InitializeModelAsync()
{
//若模型已加载，则可略过此步
if (this.model != null)
{
return;
}
//ms-appx是指当前正运行应用的程序包。应该是Assets/StreamingAssets下的文件部署至Hololens后会在Data文件夹中
StorageFile tinyYoloModelFile = await StorageFile.GetFileFromApplicationUriAsync(new Uri($"ms-appx:///Data/StreamingAssets/TinyYOLO.onnx"));
this.model = await LearningModelPreview.LoadModelFromStorageFileAsync(tinyYoloModelFile);
//评估后回收内存
this.model.InferencingOptions.ReclaimMemoryAfterEvaluation = true;
this.model.InferencingOptions.PreferredDeviceKind =
shouldUseGpu ? LearningModelDeviceKindPreview.LearningDeviceGpu : LearningModelDeviceKindPreview.LearningDeviceCpu;
//模型的输入描述是Image，输出是Tensor
var inputFeatures = this.model.Description.InputFeatures.ToArray();
var outputFeatures = this.model.Description.OutputFeatures.ToArray();
this.inputImageDescription = inputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image) as ImageVariableDescriptorPreview;
this.outputTensorDescription = outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Tensor) as TensorVariableDescriptorPreview;
if (this.model != null)
{
DisplayText("Tiny YOLO Model Loaded. Please Walk Around to Scan the Environment... ");
Debug.Log("Initialize Model Async Succeed! ");
}
}
/// <summary>
///
/// </summary>
/// <returns></returns>
private async Task LoadModelAsync()
{
if (model != null)
{
return;
}
try
{
//Outputを受け取るためには予め領域の確保が必要。
prob = new Dictionary <string, float>();
for (var i = 0; i < 1000; i++)
{
prob.Add(i.ToString(), float.NaN);
}
//Load Model
var modelFile = await StorageFile.GetFileFromApplicationUriAsync(new Uri($"ms-appx:///Assets/{ModelFileName}"));
model = await LearningModelPreview.LoadModelFromStorageFileAsync(modelFile);
// Retrieve model input and output variable descriptions (we already know the model takes an image in and outputs a tensor)
List <ILearningModelVariableDescriptorPreview> inputFeatures = model.Description.InputFeatures.ToList();
List <ILearningModelVariableDescriptorPreview> outputFeatures = model.Description.OutputFeatures.ToList();
inputImageDescription =
inputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image)
as ImageVariableDescriptorPreview;
outputMapDescription =
outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Map)
as MapVariableDescriptorPreview;
outputTensorDescription =
outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Tensor)
as TensorVariableDescriptorPreview;
//オプション設定
model.InferencingOptions.ReclaimMemoryAfterEvaluation = true;
//GPUを設定すると処理速度が上がるけれど、現時点では途中でエラーになってモデルが停止する。
//model.InferencingOptions.PreferredDeviceKind = LearningModelDeviceKindPreview.LearningDeviceGpu;
await Dispatcher.RunAsync(CoreDispatcherPriority.Normal, () => statusTBlock.Text = $"Loaded MobileNet.onnx");
}
catch (Exception ex)
{
await Dispatcher.RunAsync(CoreDispatcherPriority.Normal, () => statusTBlock.Text = $"error: {ex.Message}");
model = null;
}
}
/// <summary>
/// Load the label and model files
/// </summary>
/// <returns></returns>
private async Task LoadModelAsync()
{
await Dispatcher.RunAsync(CoreDispatcherPriority.Normal, () => StatusBlock.Text = $"Loading {_kModelFileName} ... patience ");
try
{
// Parse labels from label file
var file = await StorageFile.GetFileFromApplicationUriAsync(new Uri($"ms-appx:///Assets/{_kLabelsFileName}"));
using (var inputStream = await file.OpenReadAsync())
using (var classicStream = inputStream.AsStreamForRead())
using (var streamReader = new StreamReader(classicStream))
{
string line = "";
char[] charToTrim = { '\"', ' ' };
while (streamReader.Peek() >= 0)
{
line = streamReader.ReadLine();
line.Trim(charToTrim);
var indexAndLabel = line.Split(':');
if (indexAndLabel.Count() == 2)
{
_labels.Add(indexAndLabel[1]);
}
}
}
// Load Model
var modelFile = await StorageFile.GetFileFromApplicationUriAsync(new Uri($"ms-appx:///Assets/{_kModelFileName}"));
_model = await LearningModelPreview.LoadModelFromStorageFileAsync(modelFile);
// Retrieve model input and output variable descriptions (we already know the model takes an image in and outputs a tensor)
List <ILearningModelVariableDescriptorPreview> inputFeatures = _model.Description.InputFeatures.ToList();
List <ILearningModelVariableDescriptorPreview> outputFeatures = _model.Description.OutputFeatures.ToList();
_inputImageDescription =
inputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image)
as ImageVariableDescriptorPreview;
_outputTensorDescription =
outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Tensor)
as TensorVariableDescriptorPreview;
}
catch (Exception ex)
{
await Dispatcher.RunAsync(CoreDispatcherPriority.Normal, () => StatusBlock.Text = $"error: {ex.Message}");
_model = null;
}
}
/// <summary>
/// Load the labels and model and initialize WinML
/// </summary>
/// <returns></returns>
private async Task LoadModelAsync()
{
_evaluationLock.Wait();
{
_binding = null;
_model = null;
_isReadyForEval = false;
try
{
// Load Model
StorageFile modelFile = await StorageFile.GetFileFromApplicationUriAsync(new Uri($"ms-appx:///Assets/{_kModelFileName}.onnx"));
_model = await LearningModelPreview.LoadModelFromStorageFileAsync(modelFile);
// Hardcoding to use GPU
InferencingOptionsPreview options = _model.InferencingOptions;
options.PreferredDeviceKind = _useGPU ? LearningModelDeviceKindPreview.LearningDeviceGpu : LearningModelDeviceKindPreview.LearningDeviceCpu;
_model.InferencingOptions = options;
// Debugging logic to see the input and output of ther model
List <ILearningModelVariableDescriptorPreview> inputFeatures = _model.Description.InputFeatures.ToList();
List <ILearningModelVariableDescriptorPreview> outputFeatures = _model.Description.OutputFeatures.ToList();
var metadata = _model.Description.Metadata;
foreach (var md in metadata)
{
Debug.WriteLine($"{md.Key} | {md.Value}");
}
_inputImageDescription =
inputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image)
as ImageVariableDescriptorPreview;
_outputImageDescription =
outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image)
as ImageVariableDescriptorPreview;
_isReadyForEval = true;
}
catch (Exception ex)
{
NotifyUser($"error: {ex.Message}", NotifyType.ErrorMessage);
Debug.WriteLine($"error: {ex.Message}");
}
}
_evaluationLock.Release();
}
/// <summary>
///
/// </summary>
/// <returns></returns>
private async Task LoadModelAsysnc()
{
if (model != null)
{
return;
}
try
{
//モデルの読み込み
var modelFile = await StorageFile.GetFileFromApplicationUriAsync(new Uri($"ms-appx:///Assets/{ModelFileName}"));
model = await LearningModelPreview.LoadModelFromStorageFileAsync(modelFile);
List <ILearningModelVariableDescriptorPreview> inputFeatures = model.Description.InputFeatures.ToList();
List <ILearningModelVariableDescriptorPreview> outputFeatures = model.Description.OutputFeatures.ToList();
//入力データの形式を設定
inputImageDescription = inputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image) as ImageVariableDescriptorPreview;
//出力データの形式を設定
outputMapDescription = outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Map) as MapVariableDescriptorPreview;
outputTensorDescription = outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Tensor) as TensorVariableDescriptorPreview;
//オプション
model.InferencingOptions.ReclaimMemoryAfterEvaluation = true;
//GPUが利用できる環境ではコメントを外すと処理速度がアップする
//model.InferencingOptions.PreferredDeviceKind = LearningModelDeviceKindPreview.LearningDeviceGpu;
//Outputを受け取るためにあらかじめ記録領域を用意する必要がある。
//CustomVisionから出力したモデルの場合はTagの数だけ用意しておく。
loss = new Dictionary <string, float>();
for (var i = 0; i < lossNum; i++)
{
loss.Add(i.ToString(), float.NaN);
}
}
catch (Exception ex)
{
Debug.WriteLine(ex.Message);
}
}
private async Task LoadModelAsync()
{
try
{
// ラベルデータの読み込み
var file = await ApplicationData.Current.LocalFolder.GetFileAsync("Labels.json");
using (var inputStream = await file.OpenReadAsync())
using (var classicStream = inputStream.AsStreamForRead())
using (var streamReader = new StreamReader(classicStream))
{
string line = "";
char[] charToTrim = { '\"', ' ' };
while (streamReader.Peek() >= 0)
{
line = streamReader.ReadLine();
line.Trim(charToTrim);
var indexAndLabel = line.Split(':');
if (indexAndLabel.Count() == 2)
{
_labels.Add(indexAndLabel[1]);
}
}
}
// 学習モデルの読み込み
var modelFile = await ApplicationData.Current.LocalFolder.GetFileAsync("SqueezeNet.onnx");
_model = await LearningModelPreview.LoadModelFromStorageFileAsync(modelFile);
List <ILearningModelVariableDescriptorPreview> inputFeatures = _model.Description.InputFeatures.ToList();
List <ILearningModelVariableDescriptorPreview> outputFeatures = _model.Description.OutputFeatures.ToList();
_inputImageDescription = inputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image) as ImageVariableDescriptorPreview;
_outputTensorDescription = outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Tensor) as TensorVariableDescriptorPreview;
}
catch (Exception ex)
{
_model = null;
UnityEngine.WSA.Application.InvokeOnAppThread(() =>
{
text.text = ex.ToString();
}, true);
}
}
private async Task LoadModelAsync(bool isGpu = true)
{
if (this.model != null)
{
return;
}
await Dispatcher.RunAsync(CoreDispatcherPriority.Normal, () => StatusBlock.Text = $"Loading { MODEL_FILENAME } ... patience");
try
{
// Load Model
Debug.Write("1");
var modelFile = await StorageFile.GetFileFromApplicationUriAsync(new Uri($"ms-appx:///Assets/{ MODEL_FILENAME }"));
this.model = await LearningModelPreview.LoadModelFromStorageFileAsync(modelFile);
this.model.InferencingOptions.ReclaimMemoryAfterEvaluation = true;
this.model.InferencingOptions.PreferredDeviceKind = isGpu == true ? LearningModelDeviceKindPreview.LearningDeviceGpu : LearningModelDeviceKindPreview.LearningDeviceCpu;
Debug.Write("2");
//入力、出力を受ける変数。入力には画像、出力にテンソルが入る前提
var inputFeatures = this.model.Description.InputFeatures.ToList();
var outputFeatures = this.model.Description.OutputFeatures.ToList();
this.inputImageDescription =
inputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image) // 配列の先頭の要素を返すか、空なら規定値を返す
as ImageVariableDescriptorPreview;
this.outputTensorDescription =
outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Tensor)
as TensorVariableDescriptorPreview;
await Dispatcher.RunAsync(CoreDispatcherPriority.Normal, () => StatusBlock.Text = $"Loaded { MODEL_FILENAME }. Press the camera button to start the webcam...");
}
catch (Exception ex)
{
Debug.Write("kuso");
Debug.Write(ex.Message);
await Dispatcher.RunAsync(CoreDispatcherPriority.Normal, () => StatusBlock.Text = $"error: {ex.Message}");
model = null;
}
}
public async Task LoadModelAsync()
{
if (_isLoaded == true)
{
return;
}
//load model from assets
var modelFile = await StorageFile.GetFileFromApplicationUriAsync(new Uri(_modelUri));
_model = await LearningModelPreview.LoadModelFromStorageFileAsync(modelFile);
var inputFeatures = _model.Description.InputFeatures.ToList();
var outputFeatures = _model.Description.OutputFeatures.ToList();
//read properties out of the model for binding
_inputImageDescription = (ImageVariableDescriptorPreview)inputFeatures.FirstOrDefault();
_isLoaded = true;
}
private async Task LoadModelAsync(bool isGpu = true)
{
if (this.model != null)
{
return;
}
await Dispatcher.RunAsync(CoreDispatcherPriority.Normal, () => StatusBlock.Text = $"Loading { MODEL_FILENAME } ... patience ");
try
{
// Load Model
var modelFile = await StorageFile.GetFileFromApplicationUriAsync(new Uri($"ms-appx:///Assets/{ MODEL_FILENAME }"));
this.model = await LearningModelPreview.LoadModelFromStorageFileAsync(modelFile);
this.model.InferencingOptions.ReclaimMemoryAfterEvaluation = true;
this.model.InferencingOptions.PreferredDeviceKind = isGpu == true ? LearningModelDeviceKindPreview.LearningDeviceGpu : LearningModelDeviceKindPreview.LearningDeviceCpu;
// Retrieve model input and output variable descriptions (we already know the model takes an image in and outputs a tensor)
var inputFeatures = this.model.Description.InputFeatures.ToList();
var outputFeatures = this.model.Description.OutputFeatures.ToList();
this.inputImageDescription =
inputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image)
as ImageVariableDescriptorPreview;
this.outputTensorDescription =
outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Tensor)
as TensorVariableDescriptorPreview;
await Dispatcher.RunAsync(CoreDispatcherPriority.Normal, () => StatusBlock.Text = $"Loaded { MODEL_FILENAME }. Press the camera button to start the webcam...");
}
catch (Exception ex)
{
await Dispatcher.RunAsync(CoreDispatcherPriority.Normal, () => StatusBlock.Text = $"error: {ex.Message}");
model = null;
}
}
// Loads the ML model file and sets
private async Task LoadModelAsync()
{
if (model != null)
{
return;
}
Debug.WriteLine($"Loading Model");
try
{
// Load Model
var modelFile = await StorageFile.GetFileFromApplicationUriAsync(ModelUri);
Debug.WriteLine($"Model file discovered at: {modelFile.Path}");
model = await LearningModelPreview.LoadModelFromStorageFileAsync(modelFile);
Debug.WriteLine($"LearningModelPReview object instantiated: {modelFile.Path}");
options = model.InferencingOptions;
options.PreferredDeviceKind = LearningModelDeviceKindPreview.LearningDeviceGpu;
model.InferencingOptions = options;
// Retrieve model input and output variable descriptions (we already know the model takes an image in and outputs a tensor)
var inputFeatures = model.Description.InputFeatures.ToList();
Debug.WriteLine($"{inputFeatures.Count} Input Features");
var outputFeatures = model.Description.OutputFeatures.ToList();
Debug.WriteLine($"{inputFeatures.Count} Output Features");
inputImageDescription = inputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image)
as ImageVariableDescriptorPreview;
outputTensorDescription = outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Tensor)
as TensorVariableDescriptorPreview;
binding = new LearningModelBindingPreview(model); // Create bindings for the input and output buffer
outputArray = new List <float>(); // R4 WinML does needs the output pre-allocated for multi-dimensional tensors
outputArray.AddRange(new float[21125]); // Total size of TinyYOLO output
if (inputImageDescription != null && outputTensorDescription != null)
{
binding.Bind(inputImageDescription.Name, evaluatableVideoFrame);
binding.Bind(outputTensorDescription.Name, outputArray);
modelBindingComplete = true;
}
else
{
modelBindingComplete = false;
}
}
catch (Exception ex)
{
Debug.WriteLine($"Error loading model: {ex.Message}");
model = null;
modelBindingComplete = false;
}
}
public async Task LoadModelAsync(bool shouldUseGpu = false)
{
try
{
// Parse labels from label file
var labelsTextAsset = Resources.Load(LabelsFileName) as TextAsset;
using (var streamReader = new StringReader(labelsTextAsset.text))
{
string line = "";
char[] charToTrim = { '\"', ' ' };
while (streamReader.Peek() >= 0)
{
line = streamReader.ReadLine();
line.Trim(charToTrim);
var indexAndLabel = line.Split(':');
if (indexAndLabel.Count() == 2)
{
_labels.Add(indexAndLabel[1]);
}
}
}
#if ENABLE_WINMD_SUPPORT
// Load Model via Unity
// The WinML LearningModelPreview.LoadModelFromStreamAsync is 'not implemented' so we are using the trick to load via Unity Resource as txt file...
// Thanks Mike for the hint! https://mtaulty.com/2018/03/29/third-experiment-with-image-classification-on-windows-ml-from-uwp-on-hololens-in-unity/
IStorageFile modelFile = null;
var fileName = "model.bytes";
try
{
modelFile = await ApplicationData.Current.TemporaryFolder.GetFileAsync(fileName);
}
catch (FileNotFoundException)
{
}
if (modelFile == null)
{
var modelResource = Resources.Load(ModelFileName) as TextAsset;
modelFile = await ApplicationData.Current.TemporaryFolder.CreateFileAsync(fileName);
await FileIO.WriteBytesAsync(modelFile, modelResource.bytes);
}
// Initialize model
_model = await LearningModelPreview.LoadModelFromStorageFileAsync(modelFile);
_model.InferencingOptions.ReclaimMemoryAfterEvaluation = true;
_model.InferencingOptions.PreferredDeviceKind =
shouldUseGpu ? LearningModelDeviceKindPreview.LearningDeviceGpu : LearningModelDeviceKindPreview.LearningDeviceCpu;
// Get model input and output descriptions
List <ILearningModelVariableDescriptorPreview> inputFeatures = _model.Description.InputFeatures.ToList();
List <ILearningModelVariableDescriptorPreview> outputFeatures = _model.Description.OutputFeatures.ToList();
_inputImageDescription =
inputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image)
as ImageVariableDescriptorPreview;
_outputTensorDescription =
outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Tensor)
as TensorVariableDescriptorPreview;
#endif
}
catch
{
#if ENABLE_WINMD_SUPPORT
_model = null;
#endif
throw;
}
}
/// <summary>
/// Load the label and model files
/// </summary>
/// <returns></returns>
private async Task LoadModelAsync()
{
// just load the model one time.
if (_model != null)
{
return;
}
StatusBlock.Text = $"Loading {_kModelFileName} ... patience ";
try
{
// Parse labels from label json file. We know the file's
// entries are already sorted in order.
var fileString = File.ReadAllText($"Assets/{_kLabelsFileName}");
var fileDict = JsonConvert.DeserializeObject <Dictionary <string, string> >(fileString);
foreach (var kvp in fileDict)
{
_labels.Add(kvp.Value);
}
// legacy code for comparison
/*
* var labels = new List<string>();
* var file = await StorageFile.GetFileFromApplicationUriAsync(new Uri($"ms-appx:///Assets/{_kLabelsFileName}"));
* using (var inputStream = await file.OpenReadAsync())
* using (var classicStream = inputStream.AsStreamForRead())
* using (var streamReader = new StreamReader(classicStream))
* {
* string line = "";
* char[] charToTrim = { '\"', ' ' };
* while (streamReader.Peek() >= 0)
* {
* line = streamReader.ReadLine();
* line.Trim(charToTrim);
* var indexAndLabel = line.Split(':');
* if (indexAndLabel.Count() == 2)
* {
* labels.Add(indexAndLabel[1]);
* }
* }
* }
*/
// Load Model
var modelFile = await StorageFile.GetFileFromApplicationUriAsync(new Uri($"ms-appx:///Assets/{_kModelFileName}"));
_model = await LearningModelPreview.LoadModelFromStorageFileAsync(modelFile);
// Retrieve model input and output variable descriptions (we already know the model takes an image in and outputs a tensor)
List <ILearningModelVariableDescriptorPreview> inputFeatures = _model.Description.InputFeatures.ToList();
List <ILearningModelVariableDescriptorPreview> outputFeatures = _model.Description.OutputFeatures.ToList();
_inputImageDescription =
inputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Image)
as ImageVariableDescriptorPreview;
_outputTensorDescription =
outputFeatures.FirstOrDefault(feature => feature.ModelFeatureKind == LearningModelFeatureKindPreview.Tensor)
as TensorVariableDescriptorPreview;
}
catch (Exception ex)
{
StatusBlock.Text = $"error: {ex.Message}";
_model = null;
}
}
