This hands-on presentation will be focused on practical, essential aspects that are necessary in order to build a custom classifier. The tutorial will start from prerequisites, like the libraries that are necessary to install, to the step-by-step procedure for classifying new classes, which have not been previously learnt, by a pre-trained model using transfer learning. Such a separation of new classes of objects in images starts with the building of the novel image data set, its separation into training, validation and test sets. The model will learn to distinguish the objects from the images in the training set, it will be tuned on a validation set and finally it will face images from the previously unseen test set.
