public Task <MachineLearningResult> AnalizeImageAsync(string mlModel, Stream imageStream)
{
if (_machineLearningTask != null)
{
_machineLearningTask.TrySetCanceled();
_classifierRequestHandler.Dispose();
_classifierRequestHandler = null;
}
_machineLearningTask = new TaskCompletionSource <MachineLearningResult>();
try
{
VNImageOptions options = new VNImageOptions();
_classifierRequestHandler = new VNImageRequestHandler(NSData.FromStream(imageStream), options);
_classifierRequestHandler.Perform(new VNRequest[] { GetClassificationRequest(mlModel) }, out var err);
if (err != null)
{
Debug.WriteLine(err);
_machineLearningTask.TrySetResult(null);
}
}
catch (Exception error)
{
Debug.WriteLine(error);
_machineLearningTask.TrySetResult(null);
}
return(_machineLearningTask.Task);
}
public Task <ScanResult> ProcessImage(Stream bitmapStream)
{
var imageData = NSData.FromStream(bitmapStream);
var image = UIImage.LoadFromData(imageData);
var v = new VNImageOptions();
if (image.CGImage == null)
{
throw new Exception("No image");
}
// TODO: Find a way to make orientation foolproof
// (Probably convert stream to UIImage which has orientation encoded...)
var requestHandler =
new VNImageRequestHandler(image.CGImage, v);
var completionSource = new TaskCompletionSource <ScanResult>();
var request = new VNRecognizeTextRequest((vnRequest, error) =>
{
var results = vnRequest.GetResults <VNRecognizedTextObservation>();
var scanResult = new ScanResult();
foreach (var textObservation in results)
{
var candidate = textObservation.TopCandidates(1).FirstOrDefault();
if (candidate != null)
{
scanResult.Add(GetBlock(candidate, textObservation));
}
}
completionSource.TrySetResult(scanResult);
});
requestHandler.Perform(new VNRequest[] { request }, out var nsError);
// ReSharper disable once ConstantConditionalAccessQualifier
if (!string.IsNullOrEmpty(nsError?.Description))
{
throw new Exception(nsError.Description);
}
return(completionSource.Task);
}
public void DidOutputSampleBuffer(AVCaptureOutput captureOutput, CMSampleBuffer sampleBuffer, AVCaptureConnection connection)
{
try
{
var currentDate = DateTime.Now;
//Console.WriteLine("DidOutputSampleBuffer: " + currentDate + " " + lastAnalysis + " " + currentDate.Subtract(lastAnalysis).Milliseconds);
// control the pace of the machine vision to protect battery life
if (currentDate - lastAnalysis >= pace)
{
lastAnalysis = currentDate;
}
else
{
//Console.WriteLine("-- skip --");
return; // don't run the classifier more often than we need
}
// keep track of performance and log the frame rate
if (trackPerformance)
{
frameCount = frameCount + 1;
if (frameCount % framesPerSample == 0)
{
var diff = currentDate.Subtract(startDate);
if (diff.Seconds > 0)
{
if (pace > TimeSpan.Zero)
{
Console.WriteLine("WARNING: Frame rate of image classification is being limited by \"pace\" setting. Set to 0.0 for fastest possible rate.");
}
}
Console.WriteLine($"{diff.Seconds / framesPerSample}s per frame (average");
}
startDate = currentDate;
}
// Crop and resize the image data.
// Note, this uses a Core Image pipeline that could be appended with other pre-processing.
// If we don't want to do anything custom, we can remove this step and let the Vision framework handle
// crop and resize as long as we are careful to pass the orientation properly.
using (var croppedBuffer = CroppedSampleBuffer(sampleBuffer, targetImageSize))
{
if (croppedBuffer == null)
{
return;
}
try
{
VNImageOptions options = new VNImageOptions();
classifierRequestHandler = new VNImageRequestHandler(croppedBuffer, options);
NSError err;
classifierRequestHandler.Perform(ClassificationRequest, out err);
if (err != null)
{
Console.WriteLine(err);
}
}
catch (Exception error)
{
Console.WriteLine(error);
}
}
}
finally
{
sampleBuffer.Dispose();
}
}
