What drove the decision to develop an AI-based use case?
With increasing sales in e-commerce and the growing relevance of smartphone shopping, the aim was to create spotsize, a solution that transfers the shoe shopping experience in retail stores to online shopping. For this purpose, spotsize takes on the role of a shoe salesperson and provides customers with the right shoes. The program measures the feet using a smartphone camera and determines the appropriate shoe sizes on this basis.
However, before a corresponding program can segment the size, a 3D model of the foot must be created. The ML and AI algorithms of spotsize are modeled and trained to recognize human feet and reconstruct a 3D biometric digital twin. This allows the actual length, width, height, etc. to be calculated afterwards. A Big Data solution makes it possible to use this biometric data to complete and recommend shoe sizes more quickly. For this, patterns in the data are compared and can subsequently be used for forecasts, trends and business insights in real time.

What were your expectations or requirements for the AI application?
To recognize the feet, the user is guided by an AR user interface to take several images of the foot and calculate the appropriate shoe size. To ensure this, we had to take an investigative approach. This is because neither the 3D object classification technology used has been applied extensively in the end-user space, nor have there been many use cases with three-dimensional depth measurement of a smartphone.

What frameworks/methods did you rely on for development?
We used TensorFlow for training the deep learning models. MLFlow was used for experiment management and model management. Apple CoreML handled the conversion of the models to the smartphone and we used Jupyter notebooks for evaluation. spotsize finally ingested and labeled the datasets and built the models into the app. This was later followed by integration with SAP Cloud.

The use of AI solutions must also be accompanied by a change in competences. How have you dealt with this challenge in respect of your employees?
Due to the comparatively new technology approaches that were used, we were supposed to support spotsize in finding a suitable solution. However, despite many years of experience in the field of artificial intelligence, we also had problems with the training data, especially at the beginning. It first had to be properly coordinated what was particularly important with training data. Both inovex and our customer spotsize already had experience in the area of artificial intelligence, so that the necessary competencies were already available among the employees.

In what timeframe did they implement the AI solution in your company?
For our customer spotsize, we developed a proof of concept (PoC) in three weeks. This was followed by the integration of the solution within four months (August - November 2019).

A conclusion to your AI solution:
The exploratory approach of our AI solution for spotsize has shown that - especially when embarking on new paths in development - it is always necessary to react flexibly to technical limitations. New technologies, such as 3D object classification with smartphones may not always work smoothly with established solutions. However, once the obstacles are removed, AI can be used to tap even more potential than previously hoped for. In our case, it is now possible for the end user to capture the images of the foot via AR scan instead of taking individual static photos.

One final piece of advice/tip to other entrepreneurs looking to apply AI:
The use of artificial intelligence can significantly accelerate work steps and systems. However, the process of developing the appropriate solution can be complicated and require a high level of expertise. In order to respond quickly and flexibly to changing requirements, an agile approach to work and an interdisciplinary team offer the optimal means.

More info at:
https://spotsize.io
https://www.inovex.de