Many of us do both stills and video where stills require even better eye AF than video. Here's examples recorded via HDMI on the top current top AF cameras:
Haven't found many X-T3 latest FW examples, however here's one for video and native lenses that looks decent:
As a developer in the AI/ML field, I can tell you that the face/pattern recognition part has been well known for over 10 years. As far as computational performance, there was decent face detection running in Javascript(!) around 2012. Meaning, even C/C++ based detection on ARM cores have been possible for a very long time for embedded systems. Custom hardware/GPU based methods are orders of magnitude faster and use significantly less power (important for cameras).
When shooting stills with the Canon 1DX II, I sometimes go into LiveView so I can avoid doing Focus-Recompose however Canon only shows a face box even though I suspect they are picking the closest eye (same with the C300 II and other Canon cameras). From my experience and viewing online tests, while Canon works pretty well, perhaps second to Sony now, Canon has a limited face-size range where it can also detect the eyes. Which means it's a software/firmware issue since Canon has dual pixels and Sony (and everyone else) has far fewer phase detect pixels (and do combo phase-detect + contrast detect).
Perhaps use this thread to drop more examples of AF, eye-AF, with native, third party, and adapted lenses for future reference and to compare performance between brands, lenses, and adapters.
Haven't found many X-T3 latest FW examples, however here's one for video and native lenses that looks decent:
As a developer in the AI/ML field, I can tell you that the face/pattern recognition part has been well known for over 10 years. As far as computational performance, there was decent face detection running in Javascript(!) around 2012. Meaning, even C/C++ based detection on ARM cores have been possible for a very long time for embedded systems. Custom hardware/GPU based methods are orders of magnitude faster and use significantly less power (important for cameras).
When shooting stills with the Canon 1DX II, I sometimes go into LiveView so I can avoid doing Focus-Recompose however Canon only shows a face box even though I suspect they are picking the closest eye (same with the C300 II and other Canon cameras). From my experience and viewing online tests, while Canon works pretty well, perhaps second to Sony now, Canon has a limited face-size range where it can also detect the eyes. Which means it's a software/firmware issue since Canon has dual pixels and Sony (and everyone else) has far fewer phase detect pixels (and do combo phase-detect + contrast detect).
Perhaps use this thread to drop more examples of AF, eye-AF, with native, third party, and adapted lenses for future reference and to compare performance between brands, lenses, and adapters.