Earlier in this chapter, we pointed out that, after collecting images to be used for expression analysis and recognition, each individual image must be annotated or labelled as to which emotion group it belongs to. This can be a daunting task. Fortunately, when using IBM Watson Studio, you can simply include the appropriate images in a ZIP file and drop the ZIP file onto a class and Watson will automatically label the image file. For example, in the following screenshot, you can see that we can correctly identify those images to include within our happy class (shown outlined in green) while a single image, sad05 (shown outlined in red), does not belong to and should be excluded from our ZIP file:

This is a pretty easy process, but it could invite errors. Since it is easy and quick, you may mistakenly include images that will dilute the training sample. Keep in mind that, even if the image files are named intuitively, such as happy or sad, Watson doesn't care about the names, it simply labels all of the images in the file as positive or matching to the class.  

Finally, there's one more note about the training data. Once you go to the trouble of collecting and uploading data as an IBM Watson Studio asset, that data is available to any of your projects and, if you want to, you can share it with any other Watson Studio user! This promotes the development of assets across projects and users and increases the return on your investment.