The Smart Mirror v2 offers the possibility to promote certain products. In our first example, we first used glasses. These could be part of a new eye wear collection. In addition to the glasses, we also recommend other products to the customer that could fit to the particular type of glasses. The idea with the recommendation is from the original Smart Mirror version. We adopt it, but it could also be configurable within the Smart Mirror application. Another possibility could be that the recommendations come from the Marking Cloud. Or maybe the glasses are already offered online and Context-Driven Services (CDS) from the SAP Commerce Cloud could create the recommendations based on customer behavior.
Normally, the products that should be promoted and recognized by the Smart Mirror should be trained with different images. This could be done, for example, with SAP Leonardo Machine Learning. The glasses should ideally be trained by several people, because the environment in which the different Smart Mirrors are located is different. The creation of the images, uploading and training takes time and requires different people. In addition, the trained model must be uploaded to the Smart Mirror and replaced. With the Smart Mirror v2 we solve this problem (… ups, this sounds like advertisement).
At the moment Oktoberfest (Wiesn) is in Munich. Sven and I (Lars) came up with the idea to use suitable products and to train e.g. gingerbread hearts (Lebkuchenherzen). The recommendations for the Oktoberfest collection can be defined with an own tenant. According to the gingerbread heart different Dirndl, Lederhosen or scarves are offered. The training happens in the same way as with the glasses, where the gingerbread hearts are held in the camera. This allows new products to be trained quickly.
Normally, different images with different people, environments, brightness and viewing angles are needed for the re-training in order to get a better result and to keep bias small. The teachable machine approach used in the Smart Mirror surprisingly works very well without the need for different people to train the product. This is because the background remains the same and is independent of the person. The viewing angle when looking at the camera does not change. Likewise, the glasses are always in the same position.
The Smart Mirror is a smart prototype :)