EasyClassifyTrain

Support
Required licenses	EasyClassify
Recommended images	n/a
Location	Deep Learning Inspection/EasyClassifyTrain

Purpose

This sample program demonstrates how to:

• Train an tool in console mode.

Code highlights

By default, the sample program trains a tool for the dataset MiniWaffles in the Deep Learning Additional Resources.

Note

In the code, you need to replace DEEP_LEARNING_ADDITIONAL_RESOURCES by the path of the Deep Learning Additional Resources.

1. Select the first available GPU for the training.

   // initialize a classifier that will be trained
   EClassifier newClassifier;
   // make sure that we have a GPU to use (note: GPU's can only be detected in x64 configuration)
   std::vector<EDeepLearningDevice> devices = newClassifier.GetAvailableDevices();
   bool foundGPU = false;
   for (size_t i = 0; i < devices.size(); i++)
   {
     if (devices[i].GetDeviceType() == EDeepLearningDeviceType_GPU)
     {
       newClassifier.SetActiveDevice(devices[i]);
       foundGPU = true;
       break;
     }
   }
   

2. Create a dataset and add all the good and defective images to it.

   // load the dataset
   EClassificationDataset dataset;
   dataset.AddImages(std::string(IMAGE_SOURCE_PATH) + "mw_good*.jpg", "Good");
   dataset.AddImages(std::string(IMAGE_SOURCE_PATH) + "mw_bad*.jpg", "Bad");
   

3. Split the dataset in a training part and a validation part.

   // Put 80% of images into the training dataset and 20% in the validation dataset
   EClassificationDataset trainingDataset;
   EClassificationDataset validationDataset;
   dataset.SplitDataset(trainingDataset, validationDataset, 0.8f);
   

4. Train the classifier and wait for the training to end.

   newClassifier.Train(trainingDataset, validationDataset, 50);
   newClassifier.WaitForTrainingCompletion();
   

5. Check the validation accuracy of the classifier.

   float newClassifierAccuracy = newClassifier.GetValidationMetrics(newClassifier.GetBestIteration()).GetAccuracy();
   
   std::cout << "New classifier accuracy:" << newClassifierAccuracy << std::endl