private int ClassifyCategory(Texture2D input_tex)
{
float startclass = Time.realtimeSinceStartup;
// extract feautures from network.
var featureVector = featureExtractor.FetchOutput <List <float>, Texture2D>(input_tex);
// normalize feature vector
var output_array = GenericUtils.ConvertToDouble(featureVector.ToArray()); // convert to double format.
var norm_fv = svm_model.NormalizeElements(output_array, svm_model.muData, svm_model.sigmaData);
List <double> norm_fv_list = new List <double>(norm_fv);
// calculate propabilities
var probs = svmClassifier.FetchOutput <List <float>, List <double> >(norm_fv_list);
// get max propability class.
float maxValue = probs.Max();
int category_index = probs.IndexOf(maxValue);
float endclass = Time.realtimeSinceStartup;
classificationTime = CalculateTimeDifference(startclass, endclass);
return(category_index);
}
/// <summary>
/// This methods aims to classify a 2D texture based on the svm model that has been initialized
/// in the start method of this class.
/// </summary>
/// <param name="input_Tex"> Texture2D obj</param>
/// <returns>Integer type</returns>
private int ClassifyCategory(Texture2D input_tex)
{
Texture2D tempTex = new Texture2D(input_tex.width, input_tex.height, TextureFormat.RGB24, false);
#if UNITY_ANDROID
//Graphics.CopyTexture(input_tex, tempTex);
if (SystemInfo.copyTextureSupport == UnityEngine.Rendering.CopyTextureSupport.None)
{
Color[] sourcePixels = input_tex.GetPixels();
tempTex.SetPixels(sourcePixels);
tempTex.Apply();
}
else
{
Graphics.CopyTexture(input_tex, tempTex);
}
#endif
#if UNITY_IOS
if (SystemInfo.copyTextureSupport == UnityEngine.Rendering.CopyTextureSupport.None)
{
Color[] sourcePixels = input_tex.GetPixels();
tempTex.SetPixels(sourcePixels);
tempTex.Apply();
}
else
{
Graphics.CopyTexture(input_tex, tempTex);
}
#endif
float startclass = Time.realtimeSinceStartup;
// extract feautures from network.
var featureVector = featureExtractor.FetchOutput <List <float>, Texture2D>(tempTex);
// normalize feature vector
var output_array = ConvertToDouble(featureVector.ToArray()); // convert to double format.
var norm_fv = svm_model.NormalizeElements(output_array, svm_model.muData, svm_model.sigmaData);
List <double> norm_fv_list = new List <double>(norm_fv);
// calculate propabilities
var probs = svmClassifier.FetchOutput <List <float>, List <double> >(norm_fv_list);
// get max propability class.
float maxValue = probs.Max();
int categoryIndex = probs.IndexOf(maxValue);
float endclass = Time.realtimeSinceStartup;
classificationTime = CalculateTimeDifference(startclass, endclass);
DestroyImmediate(tempTex);
return(categoryIndex);
}
// Start is called before the first frame update
void Start()
{
TFSharpClassification classifier = new TFSharpClassification("input_1", "Logits/Softmax", 224, 224, 127.5f, 127.5f, model, labelsFile, angle, 0.05f);
var tex = Resources.Load <Texture2D>("Textures/Masoutis/" + image_file);
//var tex = Resources.Load<Texture2D>("Textures/" + image_file);
//Texture2D tex = new Texture2D(img.texture.width, img.texture.height);
//tex = img.texture as Texture2D;
RawImage img2 = GameObject.Find("img2").GetComponent <RawImage>();
img2.texture = tex;;
//var oute = classifier.FetchOutput(tex);
//foreach (KeyValuePair<string, float> value in oute)
//{
// Debug.Log("class :" + value.Key + ": " + value.Value);
//}
//print("===============================");
//IModelPrediction classifier_3 = new TFClassification("input_1", "Logits/Softmax", 224, 224, 127.5f, 127.5f, model, labelsFile, angle, 0.01f);
//var output = classifier_3.FetchOutput<IList, Texture2D>(tex);
//foreach (KeyValuePair<string, float> value in output)
//{
// Debug.Log("class :" + value.Key + ": " + value.Value);
//}
IModelPrediction classifier_4 = new TFFeatureExtraction("input_1", "block_15_project/convolution", 224, 224, 127.5f, 127.5f, model, labelsFile, angle, 0.01f);
var output = classifier_4.FetchOutput <List <float>, Texture2D>(tex);
double[] output_array = GenericUtils.ConvertToDouble(output.ToArray());
SVMClassification obj = new SVMClassification();
obj.SetModelParameters("SVM_Weights", "mu", "sigma");
var norm_fv = obj.NormalizeElements(output_array, obj.muData, obj.sigmaData);
List <double> norm_fv_list = new List <double>(norm_fv);
IModelPrediction svm = obj;
var probs = svm.FetchOutput <List <float>, List <double> >(norm_fv_list);
foreach (var item in probs)
{
Debug.Log(item);
}
}