We have now created a simple biometric authentication proof of concept using IBM Watson Studio and the IBM Watson Visual Recognition service. It works but of course this is not ready for deployment. There are many things to consider before a model is ready for deployment and production use. For example, guidelines provided with the Visual Recognition service state that it is best practice to include at least 50 positive images per class before you can begin to realistically assess your training results.

Other recommendations provided include the following:

• Assuming similar quality and content for your training data, more training images generally provide more accurate results than fewer images.
• 150 to 200 images per .zip file provides the best balance between processing time and accuracy. More than 200 images will increase the time and the accuracy, but with diminishing returns for the amount of time it takes.