Methods and means for real-time object recognition accuracy increase in video images on iOS mobile platform

As a result of the analytical review, it was established that the family of Yolo models is a promising area of search and recognition of objects. However, existing implementations do not support the ability to run the model on the iOS platform. To achieve these goals, a comprehensive scalable conversion system has been developed to improve the recognition accuracy of arbitrary models based on the Docker system. The method of improvement is to add a layer with the Mish activation function to the original model. The method of conversion is to quickly convert any Yolo model to CoreML format.