In traditional algorithms, the developer’s knowledge gets implemented as a program, which analyzes some input data to evaluate a result. Artificial Intelligence (AI) and particularly its sub-field, machine learning (ML) establish a way to let the machine to learn by itself, based on a set of input data and known results. After this learning (or training) process, the machine can to analyze the input data and suggest a result, based on the knowledge, gathered during the training. The developer does not need to understand and decide how exactly the algorithm works. One just must know how to train the machine.
Theoretically, ML algorithms can achieve better results than a human-designed algorithm, at the expense of much higher computational power. Contemporary processors already have the necessary power, which enables ML algorithms to be used even for image processing.
The most popular field where AI/ML algorithms prove to be much better than their predecessors is image classification and objects detection in images. Using AI/ML as object detectors is a key milestone in reaching the human brain power of automated scene analysis and taking the consequent decisions.
Can run on a GPU or DSP to save power.
The neural network can be re-trained to be customer-specific instead of people-specific.
The background can be replaced or blurred, creating a variety of effects: single camera Bokeh/ Background blur, black-and-white background, background replacement, etc.
Bokeh / Background Blur
Black and White
If you are considering introducing our technology or products, or if you are interested in co-creation with us, please feel free to contact us from here.