public async Task Classify(List <List <float> > dataframe)
{
var mappedByteBuffer = GetModelAsMappedByteBuffer();
var interpreter = new Xamarin.TensorFlow.Lite.Interpreter(mappedByteBuffer);
var tensor = interpreter.GetInputTensor(0);
var shape = tensor.Shape();
var width = shape[1];
var height = shape[2];
var sr = new StreamReader(Application.Context.Assets.Open("labels.txt"));
var labels = sr.ReadToEnd().Split('\n').Select(s => s.Trim()).Where(s => !string.IsNullOrEmpty(s)).ToList();
var outputLocations = new float[1][] { new float[labels.Count] };
var outputs = Java.Lang.Object.FromArray(outputLocations);
//interpreter.Run(dataframe, outputs);
var classificationResult = outputs.ToArray <float[]>();
}
public void Detect(byte[] image)
{
var mappedByteBuffer = GetModelAsMappedByteBuffer();
var interpreter = new Xamarin.TensorFlow.Lite.Interpreter(mappedByteBuffer);
var tensor = interpreter.GetInputTensor(0);
var tensor2 = interpreter.GetOutputTensor(0);
var shape1 = tensor2.Shape();
var shape = tensor.Shape();
var width = shape[1];
var height = shape[2];
var byteBuffer = GetPhotoAsByteBuffer(image, width, height);
var streamReader = new StreamReader(Android.App.Application.Context.Assets.Open("labels.txt"));
var labels = streamReader.ReadToEnd().Split('\n').Select(s => s.Trim()).Where(s => !string.IsNullOrEmpty(s)).ToList();
//Convert our two-dimensional array into a Java.Lang.Object, the required input for Xamarin.TensorFlow.List.Interpreter
var outputLocations = new float[1, 13, 13, 50];
var output = ByteBuffer.AllocateDirect(33800);
try
{
interpreter.Run(byteBuffer, output);
}
catch (Exception ex)
{
var x = ex.Message;
}
}
public async Task Classify(byte[] bytes)
{
var modelByteBuffer = LoadModel("dog-identification.tflite");
var interpreter = new Xamarin.TensorFlow.Lite.Interpreter(modelByteBuffer);
var inputTensor = interpreter.GetInputTensor(0);
var shape = inputTensor.Shape();
var width = shape[1];
var height = shape[2];
List <String> labels = new List <String>();
var labelsStream = Application.Context.Assets.Open("labels.txt");
using (StreamReader sr = new StreamReader(labelsStream))
{
string line;
while ((line = sr.ReadLine()) != null)
{
labels.Add(line);
}
}
var input = GetPhotoAsByteBuffer(bytes, width, height);
var outputLocations = new float[1][] { new float[labels.Count] };
var outputs = Java.Lang.Object.FromArray(outputLocations);
interpreter.Run(input, outputs);
var classificationResults = outputs.ToArray <float[]>();
System.Console.WriteLine("I got results: ");
for (int i = 0; i < classificationResults[0].Length; i++)
{
System.Console.WriteLine($"{labels[i]} : {classificationResults[0][i]}");
}
var results = new List <Classification>();
for (var i = 0; i < labels.Count; i++)
{
var label = labels[i];
results.Add(new Classification(label, classificationResults[0][i]));
}
ClassificationCompleted?.Invoke(this, new ClassificationEventArgs(results));
}
public async Task Classify(byte[] bytes)
{
var mappedByteBuffer = GetModelAsMappedByteBuffer();
// The following line with fail with compilation error
// Error CS0234 The type or namespace name 'GPU' does not exist in the namespace 'Xamarin.TensorFlow.Lite' (are you missing an assembly reference?)
var gpuDelegate = new Xamarin.TensorFlow.Lite.GPU.GpuDelegate();
Xamarin.TensorFlow.Lite.Interpreter.Options options = new Xamarin.TensorFlow.Lite.Interpreter.Options().AddDelegate(gpuDelegate);
var interpreter = new Xamarin.TensorFlow.Lite.Interpreter(mappedByteBuffer, options);
// If we comment 3 lines above and uncomment 1 line below everything works fine.
//var interpreter = new Xamarin.TensorFlow.Lite.Interpreter(mappedByteBuffer);
var tensor = interpreter.GetInputTensor(0);
var shape = tensor.Shape();
var width = shape[1];
var height = shape[2];
var byteBuffer = GetPhotoAsByteBuffer(bytes, width, height);
var sr = new StreamReader(Application.Context.Assets.Open("labels.txt"));
var labels = sr.ReadToEnd().Split('\n').Select(s => s.Trim()).Where(s => !string.IsNullOrEmpty(s)).ToList();
var outputLocations = new float[1][] { new float[labels.Count] };
var outputs = Java.Lang.Object.FromArray(outputLocations);
interpreter.Run(byteBuffer, outputs);
var classificationResult = outputs.ToArray <float[]>();
var result = new List <Classification>();
for (var i = 0; i < labels.Count; i++)
{
var label = labels[i];
result.Add(new Classification(label, classificationResult[0][i]));
}
ClassificationCompleted?.Invoke(this, new ClassificationEventArgs(result));
}
public List <ImageClassificationModel> Classify(byte[] image)
{
MappedByteBuffer mappedByteBuffer = GetModelAsMappedByteBuffer();
// TODO: Find a solution that is not being deprecated.
Xamarin.TensorFlow.Lite.Interpreter interpreter = new Xamarin.TensorFlow.Lite.Interpreter(mappedByteBuffer);
//To resize the image, we first need to get its required width and height
Xamarin.TensorFlow.Lite.Tensor tensor = interpreter.GetInputTensor(0);
int[] shape = tensor.Shape();
int width = shape[1];
int height = shape[2];
ByteBuffer byteBuffer = GetPhotoAsByteBuffer(image, width, height);
//use StreamReader to import the labels from labels.txt
StreamReader streamReader = new StreamReader(Application.Context.Assets.Open("labels.txt"));
//Transform labels.txt into List<string>
List <string> labels = streamReader.ReadToEnd().Split('\n').Select(s => s.Trim()).Where(s => !string.IsNullOrEmpty(s)).ToList();
//Convert our two-dimensional array into a Java.Lang.Object, the required input for Xamarin.TensorFlow.List.Interpreter
float[][] outputLocations = new float[1][] { new float[labels.Count] };
Java.Lang.Object outputs = Java.Lang.Object.FromArray(outputLocations);
interpreter.Run(byteBuffer, outputs);
float[][] classificationResult = outputs.ToArray <float[]>();
//Map the classificationResult to the labels and sort the result to find which label has the highest probability
List <ImageClassificationModel> classificationModelList = new List <ImageClassificationModel>();
for (int i = 0; i < labels.Count; i++)
{
string label = labels[i]; classificationModelList.Add(new ImageClassificationModel(label, classificationResult[0][i]));
}
return(classificationModelList);
}
public void Classify(byte[] bytes)
{
var mappedByteBuffer = GetModelAsMappedByteBuffer();
var interpreter = new Xamarin.TensorFlow.Lite.Interpreter(mappedByteBuffer);
var tensor = interpreter.GetInputTensor(0);
var shape = tensor.Shape();
var width = shape[1];
var height = shape[2];
var sr = new StreamReader(Application.Context.Assets.Open("hotdog-or-not-labels.txt"));
var labels = sr.ReadToEnd().Split('\n').Select(s => s.Trim()).Where(s => !string.IsNullOrEmpty(s)).ToList();
var outputLocations = new float[1][] { new float[labels.Count] };
var outputs = Java.Lang.Object.FromArray(outputLocations);
var byteBuffer = GetPhotoAsByteBuffer(bytes, width, height);
interpreter.Run(byteBuffer, outputs);
var predictionResult = outputs.ToArray <float[]>();
var result = new Dictionary <string, float>();
for (var i = 0; i < labels.Count; i++)
{
var label = labels[i];
result.Add(label, predictionResult[0][i]);
}
ClassificationCompleted?.Invoke(this, new ClassificationEventArgs(result));
}
public List <string> PredictNextWord(string word)
{
switch (Language)
{
case "pt":
using (Stream ms = context.Assets.Open("Models/idx2word_pt.txt"))
{
using (var sr = new StreamReader(ms))
{
idx2word = Newtonsoft.Json.JsonConvert.DeserializeObject <Dictionary <int, String> >(sr.ReadToEnd());
}
}
using (Stream ms = context.Assets.Open("Models/word2idx_pt.txt"))
{
using (var sr = new StreamReader(ms))
{
word2idx = Newtonsoft.Json.JsonConvert.DeserializeObject <Dictionary <String, int> >(sr.ReadToEnd());
}
}
if (!word2idx.ContainsKey(word))
{
return(new List <string>());
}
var il = new float[10];
for (var i = 0; i < 10; i++)
{
il[i] = (float)(word2idx[word]);
}
if (model == null)
{
var assets = Application.Context.Assets;
AssetFileDescriptor fileDescriptor = assets.OpenFd("Models/predictor_pt.tflite");
FileInputStream inputStream = new FileInputStream(fileDescriptor.FileDescriptor);
FileChannel fileChannel = inputStream.Channel;
long startOffset = fileDescriptor.StartOffset;
long declaredLength = fileDescriptor.DeclaredLength;
var asd = fileChannel.Map(FileChannel.MapMode.ReadOnly, startOffset, declaredLength);
model = new Xamarin.TensorFlow.Lite.Interpreter(asd);
}
byte[] ibytes = new byte[il.Length * sizeof(float)];
Buffer.BlockCopy(il, 0, ibytes, 0, ibytes.Length);
var bytebuffer = Java.Nio.ByteBuffer.Wrap(ibytes);
var output = Java.Nio.ByteBuffer.AllocateDirect(4 * 10 * (idx2word.Count + 1));
model.Run(bytebuffer, output);
List <Double> res = new List <double>();
var buffer = new byte[4 * 10 * (idx2word.Count + 1)];
Marshal.Copy(output.GetDirectBufferAddress(), buffer, 0, 4 * 10 * (idx2word.Count + 1));
for (var i = 0; i < idx2word.Count + 1; i++)
{
res.Add(BitConverter.ToSingle(buffer, i * 4));
}
return(res.OrderByDescending(x => x).Select(x2 => idx2word[res.IndexOf(x2)]).Take(10).ToList());
case "es":
using (Stream ms = context.Assets.Open("Models/idx2word_es.txt"))
{
using (var sr = new StreamReader(ms))
{
idx2word = Newtonsoft.Json.JsonConvert.DeserializeObject <Dictionary <int, String> >(sr.ReadToEnd());
}
}
using (Stream ms = context.Assets.Open("Models/word2idx_es.txt"))
{
using (var sr = new StreamReader(ms))
{
word2idx = Newtonsoft.Json.JsonConvert.DeserializeObject <Dictionary <String, int> >(sr.ReadToEnd());
}
}
if (!word2idx.ContainsKey(word))
{
return(new List <string>());
}
var il2 = new float[10];
for (var i = 0; i < 10; i++)
{
il2[i] = (float)(word2idx[word]);
}
if (model == null)
{
var assets = Application.Context.Assets;
AssetFileDescriptor fileDescriptor = assets.OpenFd("Models/predictor_es.tflite");
FileInputStream inputStream = new FileInputStream(fileDescriptor.FileDescriptor);
FileChannel fileChannel = inputStream.Channel;
long startOffset = fileDescriptor.StartOffset;
long declaredLength = fileDescriptor.DeclaredLength;
var asd = fileChannel.Map(FileChannel.MapMode.ReadOnly, startOffset, declaredLength);
model = new Xamarin.TensorFlow.Lite.Interpreter(asd);
}
byte[] ibytes2 = new byte[il2.Length * sizeof(float)];
Buffer.BlockCopy(il2, 0, ibytes2, 0, ibytes2.Length);
var bytebuffer2 = Java.Nio.ByteBuffer.Wrap(ibytes2);
var output2 = Java.Nio.ByteBuffer.AllocateDirect(4 * 10 * (idx2word.Count + 1));
model.Run(bytebuffer2, output2);
List <Double> res2 = new List <double>();
var buffer2 = new byte[4 * 10 * (idx2word.Count + 1)];
Marshal.Copy(output2.GetDirectBufferAddress(), buffer2, 0, 4 * 10 * (idx2word.Count + 1));
for (var i = 0; i < idx2word.Count + 1; i++)
{
res2.Add(BitConverter.ToSingle(buffer2, i * 4));
}
return(res2.OrderByDescending(x => x).Select(x2 => idx2word[res2.IndexOf(x2)]).Take(10).ToList());
default:
using (Stream ms = context.Assets.Open("Models/idx2word_en.txt"))
{
using (var sr = new StreamReader(ms))
{
idx2word = Newtonsoft.Json.JsonConvert.DeserializeObject <Dictionary <int, String> >(sr.ReadToEnd());
}
}
using (Stream ms = context.Assets.Open("Models/word2idx_en.txt"))
{
using (var sr = new StreamReader(ms))
{
word2idx = Newtonsoft.Json.JsonConvert.DeserializeObject <Dictionary <String, int> >(sr.ReadToEnd());
}
}
if (!word2idx.ContainsKey(word))
{
return(new List <string>());
}
var il3 = new float[10];
for (var i = 0; i < 10; i++)
{
il3[i] = (float)(word2idx[word]);
System.Diagnostics.Debug.WriteLine(il3[i]);
}
if (model == null)
{
var assets = Application.Context.Assets;
AssetFileDescriptor fileDescriptor = assets.OpenFd("Models/predictor_en.tflite");
FileInputStream inputStream = new FileInputStream(fileDescriptor.FileDescriptor);
FileChannel fileChannel = inputStream.Channel;
long startOffset = fileDescriptor.StartOffset;
long declaredLength = fileDescriptor.DeclaredLength;
var asd = fileChannel.Map(FileChannel.MapMode.ReadOnly, startOffset, declaredLength);
model = new Xamarin.TensorFlow.Lite.Interpreter(asd);
}
byte[] ibytes3 = new byte[il3.Length * sizeof(float)];
Buffer.BlockCopy(il3, 0, ibytes3, 0, ibytes3.Length);
var bytebuffer3 = Java.Nio.ByteBuffer.Wrap(ibytes3);
var output3 = Java.Nio.ByteBuffer.AllocateDirect(4 * 10 * (idx2word.Count + 1));
model.Run(bytebuffer3, output3);
List <Double> res3 = new List <double>();
var buffer3 = new byte[4 * 10 * (idx2word.Count + 1)];
Marshal.Copy(output3.GetDirectBufferAddress(), buffer3, 0, 4 * 10 * (idx2word.Count + 1));
for (var i = 0; i < idx2word.Count + 1; i++)
{
res3.Add(BitConverter.ToSingle(buffer3, i * 4));
}
return(res3.OrderByDescending(x => x).Select(x2 => idx2word[res3.IndexOf(x2)]).Take(10).ToList());
}
}
