How can i give a trained model some images as an input ?

I trained a model with classification learner to classify fruits. Now i want to test my model. How can i give the model some images as an input ?

Answer 1

[Mukul Rao]

Hello, in order to use the trained model to predict new data, you will first have to export the trained model to the workspace. You can click on the "Export Model" drop-down in the Classification Learner App, and select the first option. Once you have the trained model in the MATLAB workspace, you can call its "predict" method and supply new image data to make predictions.

[Lukas Sohlbach]

Okay, i get it so far. As i understand needs the predict method a table as input. And the input must match the original training data. How do i convert an single image into a table and match it to my training data because to train my model i used more then one image? I attached the code which i used to train my model.

imset = imageSet ('Fruits','recursive');
bag = bagOfFeatures(imset,'VocabularySize',200,'PointSelection','Detector');
imagefeatures = encode(bag,imset);
FruitData = array2table(imagefeatures);
FruitData.flagType = getImageLabels(imset);

The table "FruitData" is the input for the classification learner. I hope i was able to express my problem/question understandable.

[Mukul Rao]

Hi Lukas, yes that makes sense. You might want to take a look at this example that demonstrates the workflow. Using the bag of features, you will have to create an "imageCategoryClassifierObject" with the "trainImageCategoryClassifier" function. You can call the "predict" method on the classifier object and plug in the the new image data set as the input without having to convert it to a table. I believe you might have referred the predict method for a different classifier class in the documentation.

[Lukas Sohlbach]

Hi Mark, does the "imageCategoryClassifierObject" only works on SVM ? Because my trained model based on a Decision Tree. I attached my trained classifier. This classifier has a function "c.predictFcn" link (where c is a variable for the classifier name) and this function requests a table as an input. Is there any possibility, after using a bagOfFeatures to train a Decision Tree, to this function to make a prediction on a new image? By the way, thanks for helping me out.