In this chapter will look into two other types of Computer Vision Problems:
The first one occurs when you want to predict more than one label per image, for instance predict a dog and a cat in a same image.
The second one occurs when the labels are one or several numbers, a quantity instead of a category. And will also learn about output activations , targets and loss functions in DL models
Multi - Label Classification refers to a problem of identifying the categories of objects in a images that may not contain exactly same type of object, or there may be no objects at all in the class we are looking for.
For instance so far in our Image Classifier, whenever we input a image which is not of the classes our model would still point out that belongs to the same class with low confidence. For instance in Dog Bree Classifer what if you input a picture of horse?
Well our model will return that its some kinda of dog breed but in reality we do know that isn't. But with multi - label classification we can solve this issue with our classifiers!
In practice it's common to have some images with the zero matches or more than one match, we should probably expect in practice multi - label classifiers are more widely applicable than single label classifiers.
Here the architecture remains the same but our loss function would change.
In this example we are going to use the PASCAL dataset, which can have more than one kind of classifed object per image.
This data would be different from the ones we have seen before, in that it's not structured by filename or folder but instead comes with a CSV telling us the labels to use for each image
We have a DataFrame object how do we convert this into a DataLoaders object? We usually build a datablock and with that create a DataLoaders object. Here in this tutorial will go step by step to build a DataBlock!
A collection that returns a tuple of the independet and dependent variable for a single item.