public async Task <IMachineLearningOutput> EvaluateAsync(IMachineLearningInput input)
{
var modelInput = input as TinyYOLOModelModelInput;
TinyYOLOModelModelOutput output = new TinyYOLOModelModelOutput();
LearningModelBindingPreview binding = new LearningModelBindingPreview(LearningModel);
binding.Bind("image", modelInput.image);
binding.Bind("grid", output.grid);
LearningModelEvaluationResultPreview evalResult = await LearningModel.EvaluateAsync(binding, string.Empty);
return(output);
}
public async Task <IMachineLearningOutput> EvaluateAsync(IMachineLearningInput input)
{
var modelInput = input as TinyYOLOInput;
Binding.Bind("image", modelInput.image);
var result = await Session.EvaluateAsync(Binding, "0");
var output = new TinyYOLOOutput();
output.grid = result.Outputs["grid"] as TensorFloat16Bit;
return(output);
}
public async Task <IMachineLearningOutput> EvaluateAsync(IMachineLearningInput input)
{
var modelInput = input as FNSInput;
Binding.Bind("inputImage", modelInput.inputImage);
var result = await Session.EvaluateAsync(Binding, "0");
var output = new FNSOutput();
output.outputImage = result.Outputs["outputImage"] as TensorFloat16Bit;
return(output);
}
public async Task <IMachineLearningOutput> EvaluateAsync(IMachineLearningInput input)
{
var modelInput = input as Inceptionv3Input;
Binding.Bind("image", modelInput.image);
var result = await Session.EvaluateAsync(Binding, "0");
var output = new Inceptionv3Output();
output.classLabel = result.Outputs["classLabel"] as TensorString;
output.classLabelProbs = result.Outputs["classLabelProbs"] as IList <Dictionary <string, float> >;
return(output);
}
public async Task <IMachineLearningOutput> EvaluateAsync(IMachineLearningInput input)
{
var modelInput = input as GoogLeNetPlacesModelModelInput;
GoogLeNetPlacesModelModelOutput output = new GoogLeNetPlacesModelModelOutput();
LearningModelBindingPreview binding = new LearningModelBindingPreview(LearningModel);
binding.Bind("sceneImage", modelInput.sceneImage);
binding.Bind("sceneLabel", output.sceneLabel);
binding.Bind("sceneLabelProbs", output.sceneLabelProbs);
LearningModelEvaluationResultPreview evalResult = await LearningModel.EvaluateAsync(binding, string.Empty);
return(output);
}
public async Task <IMachineLearningOutput> EvaluateAsync(IMachineLearningInput input)
{
var modelInput = input as GoogLeNetPlacesInput;
Binding.Bind("sceneImage", modelInput.sceneImage);
var result = await Session.EvaluateAsync(Binding, "0");
var output = new GoogLeNetPlacesOutput();
output.sceneLabel = result.Outputs["sceneLabel"] as TensorString;
output.sceneLabelProbs = result.Outputs["sceneLabelProbs"] as IList <Dictionary <string, float> >;
return(output);
}
public async Task <IMachineLearningOutput> EvaluateAsync(IMachineLearningInput input)
{
var modelInput = input as ResNet50ModelInput;
ResNet50ModelOutput output = new ResNet50ModelOutput();
LearningModelBindingPreview binding = new LearningModelBindingPreview(LearningModel);
binding.Bind("image", modelInput.image);
binding.Bind("classLabel", output.classLabel);
binding.Bind("classLabelProbs", output.classLabelProbs);
LearningModelEvaluationResultPreview evalResult = await LearningModel.EvaluateAsync(binding, string.Empty);
return(output);
}
public async Task <IMachineLearningOutput> EvaluateAsync(IMachineLearningInput input)
{
var modelInput = input as RN1015k500Input;
Binding.Bind("data", modelInput.data);
var result = await Session.EvaluateAsync(Binding, "0");
var output = new RN1015k500Output();
output.classLabel = result.Outputs["classLabel"] as TensorString;
output.softmax_output = result.Outputs["softmax_output"] as IList <Dictionary <string, float> >;
return(output);
}
public async Task <IMachineLearningOutput> EvaluateAsync(IMachineLearningInput input)
{
if (input is SqueezeNetInput modelInput)
{
Binding.Bind("data_0", modelInput.Image);
}
var results = await Session.EvaluateAsync(Binding, "0");
if (!(results.Outputs["softmaxout_1"] is TensorFloat resultTensor))
{
return(null);
}
var resultVector = resultTensor.GetAsVectorView();
// Find the top 3 probabilities
var indexedResults = resultVector.Select((t, i) => (index: i, probability: resultVector.ElementAt(i)))
.ToList();
indexedResults.Sort((a, b) => {
if (a.probability < b.probability)
{
return(1);
}
if (a.probability > b.probability)
{
return(-1);
}
return(0);
});
var ret = new List <(string label, float probability)>();
for (var i = 0; i < 3; i++)
{
ret.Add((_labels[indexedResults[i].index], indexedResults[i].probability));
}
var output = new SqueezeNetOutput {
classLabelProbs = ret
};
return(output);
}
/// <summary>
/// evaluate a model
/// </summary>
/// <param name="input"></param>
/// <param name="learningModel"></param>
/// <returns></returns>
public async static Task <(IMachineLearningOutput Result, double EvalutionTime)> EvaluateWithTimeReturnAsync(IMachineLearningInput input, IMachineLearningModel learningModel)
{
var startTime = DateTime.Now;
var output = await EvaluateAsync(input, learningModel);
var costTime = (DateTime.Now - startTime).TotalSeconds;
return(output, costTime);
}
/// <summary>
/// evaluate a model
/// </summary>
/// <param name="input"></param>
/// <param name="learningModel"></param>
/// <returns></returns>
public async static Task <IMachineLearningOutput> EvaluateAsync(IMachineLearningInput input, IMachineLearningModel learningModel)
{
var output = await learningModel.EvaluateAsync(input);
return(output);
}
