(Music) In this video, we will talk about training image classifiers using IBM Watson Visual Recognition service and use the training image classifiers to classify images. Specifically, we'll cover what are custom classifiers? How can we train custom classifiers? What are the positive and negative examples for training the custom classifiers? If you want to be able to use custom labels, for specific types of objects that are not included in the pre-trained classifiers, then what you might want to do, is create your own custom classifier. For example, if you wanted to classify different car makes like Porsche versus BMW, then you won't be able to use a pre-trained classifier since it wasn't trained to identify these brands. You will be able to achieve this with custom classifiers. So how do you train custom classifiers? The concept behind the custom classifiers is really simple, and IBM Watson actually does all the work for you. All you really need, are images of what you'd like to classify and IBM Watson will do all the deep learning and training for you behind the scenes. Let's suppose, you want to create a new custom classifier to classify fruits like apples, bananas, and pears. So you declare the three labels or classes you want your custom classifier to recognize which are apples, bananas and pears. Next, you gather numerous images of apples, numerous images of bananas, and numerous images of pears. Of course, the more images per class, the better the classifier can learn from your examples and even differentiate between the classes. Especially, since you might have green apples that might look like green pears in some of your images. Note, that there is a minimum number of images per class needed to train a custom classifier with IBM Watson. You use at least 10 images for each class or you can check the documentation. The images of apples, bananas, and pears are called positive examples because they are the object that you are interested in and you have labels that uniquely identify each class. Then, after loading your images in their appropriate class, you can start training your classifier. You'll need to wait a few minutes while the service train the model and build the custom classifier. This part is all done by IBM Watson behind the scenes which makes it particularly convenient if you don't have the time or resources to understand and build your own deep learning model. The advantage of using IBM Watson Image Recognition service, is that all of your responsibility is to prepare are the images. After training completes, you can now use your custom classifier and test using images in our test set. The test set is the set of images that you didn't use to train your classifier. Note, that your classifier only available to you, meaning that no one else can see your classifier unless you give them the credentials to your classifier. As you can see, for each of the test images, it does a decent job at classifying the fruits correctly. For the image of the pear, it correctly identifies pears with a confidence of 0.76. But note, that it also has a 0.43 confidence for the apples label meaning that if you want to increase the confidence distance between pears and apples, the difference between the confidence level for the pears and apples, you might want to add more examples of pears and apple when training your classifier. For bananas, it correctly identifies bananas with a high confidence of 0.90 and it also does a great job at giving low confidences for the labels apples and pears. Similar to the banana image, the classifier does a great job at correctly classifying apples as apples. Across these three test images, you can see that the custom classifier correctly labeled each image, if you were to just take the label with the highest confidence score. In practice, depending on your situation, you may only care about this most confident label. But it's also good to note where your labels might have a good chance of getting labels mixed up, as it might be the case here with pears and apples. Thank you for watching this video. (Music)