Deep learning is the term Canon uses to describe how its AI-based autofocus subject recognition system has been taught, using millions of reference images to create a data library, which in turn is used to improve focusing performance for certain subjects.
In October 2019, when the development of the 1D X Mark III was announced, Canon stated that the camera would benefit from ‘deep learning’. This isn’t perhaps AI in the traditional sense, as it’s not on-the-job learning taking place in-camera, but the new EOS iTR AFX algorithm is the result of machine learning, with millions of reference images used. By creating a data library using sample images, the camera can then check the scene against this data to detect the type of subject being shot.
There were two libraries of data developed for the EOS-1D X Mark III – one for the viewfinder AF system, which uses EOS iTR AF, and the other for Live View.
It is the Live View sensor-based focusing system that’s being further developed in the R-series mirrorless cameras. The EOS R5 and R6 gained Animal AF, whilst the EOS R3 – due for release in the second half of 2021 – will also be able to detect and track vehicles such as cars and motorbikes.
The library of data achieved via this deep learning technology is stored and powered by Canon DIGIC X generation of processors.
Related terms