End to end implementation of machine learning solutions involves multiple sub-steps like:
• Model generation
• Data clean-up
• Model training and validation
For embedded application-based inference models, the final code needs to be implemented in the C/C++. This makes it easy to cross-compile since most of the embedded platform support C/C++ compilers. The entire setup needs a team of 4 to 5 people with a background in python, C++, embedded systems, and machine learning framework building and optimization.
Small scale companies cannot afford this kind of investment. The product was used to automate all the complicated operations to expedite the testing of the model performance. All that is required is from the user is the dataset. The entire model building, model optimization and C++ conversion as well as inference testing on the target hardware can be performed using a simple UI application.
1. Any person with a basic idea of the machine learning implementation can create and test models without writing a single line of code.
2. Inception of model creation to first test on embedded hardware within hours.