Famateur said:
"You keep using that word
inconceivable innovation. I do not think it means what you think it means."
If you mean
low ISO dynamic range, huge megapixel counts or a radical new sensor design, okay. In my book, though, DPAF is clearly in the "innovation" category.
It's certainly an interesting approach, but in my mind, innovation is more than just doing something in a new way. Innovation is defined by the "aha" moment where the advantages to the new approach are immediately so obvious that you can't imagine having done it the old way.
The thing is, Fuji showed the first compact camera with on-die phase detection AF back in 2010. Canon has been playing catch-up ever since. Yes, DPAF has an advantage over dedicated focus pixels in that you don't lose the light that would otherwise fall on half of certain pixels, though that difference will matter less and less as resolution increases—but DPAF still feels more like a way to work around Fuji's on-sensor phase detection patents while still achieving the same benefits, rather than true innovation.
If Canon wants folks like me to see them as still innovating in the area of sensors, they should:
- Start with a backside illumination design.
- Etch both sides of the sensor, with vias for every pixel.
- Put per-pixel buffers on the reverse side of the sensor die, thus giving you a true global shutter.
- Put a sizable heat sink on the reverse side of the sensor to dissipate the heat from the back-side buffers, thus reducing thermal noise.
- Use one or more on-die ADC circuits for maximum accuracy and minimum noise.
- Take advantage of the global shutter to eliminate dynamic range limitations and remove the need for setting ISO values entirely.
That last one is the jaw dropper. The benefits of a global shutter for video are obvious. The benefits for stills are even bigger, though, and I don't think anybody is really taking advantage of that yet, which seems bizarre to me.
The entire reason image sensors have limited dynamic range is twofold: because the ADC can provide only a certain number of bits of precision, and because when the full-well capacity of a pixel is exceeded, that pixel cannot hold any more photons. However, if you can get the read noise levels low enough, you can just sample the pixels several times per exposure, and sum the results in a wider register. You can then make clever use of Huffman coding or run-length encoding to minimize the impact of all those extra zeroes, and you'll be able to accurately reproduce everything from a single photon all the way up to the brightest light.
That would be innovation. Real innovation changes things in ways that are jaw-dropping and earth-shattering. Using DPAF to do automatic AFMA might do that, and using DPAF to correct the slightly-off phase detect focusing results after the mirror goes up might do that, but DPAF by itself doesn't do that, IMO. DPAF is clever, but it is far from pushing the limits of technology.
Just my $0.02.
