1DX III vs R5 AF

[Joel Eade]

This really could also be titled phase detect AF vs dual pixel AF.

So one of the things that surprised the socks off of me was that normal AF was far superior to dual pixel in even moderate low light, normal indoor lighting at night. And in dark situations it wasn't even funny the difference.

In good lighting dual pixel AF was pretty good. But I found traditional AF still more accurate and consistent especially with razor thin DOF of the105mm f/1.4 and the 50mm f/1.2.

So my question is the R5 AF better compared to the 1DX III's dual pixel AF? Because from what everyone was saying the 1DX III dual pixel AF reached mirrorless level and was able to be used in both capacities. And I have yet to hear a bad thing about mirrorless AF.

Now the eye detect was amazing to use, but not accurate enough for me to trust it for an important shoot. At least not without lots of chimping. Granted I am using very thin DOF/challenging lenses and I do realize that it's not the same as focusing for a landscape, but found that there was a big difference in accuracy between the two AF systems.

Has anyone else experienced this? Because if this is the AF experience of mirrorless I can't say it is mature enough to replace DSLR's just yet. At least not for me personally.

I guess I am a bit confused about this issue. Did you take photos with a 1DXIII and an R5 when you noticed this difference or was it some other camera compared to the R5 or 1DXIII?

I think dual pixel AF has been around at least 6 or 7 years and is used on many Canon models but I'm unsure which ones have it as opposed to the "traditional" phase detect AF.

My understanding is that cameras with "traditional phase detect AF" use separate (smaller) areas of the sensor for phase aligning focus and other areas for image capture such that all pixels are not involved in focusing. Dual pixel autofocus is sort of a variant of phase detect where each pixel has a dual diode with a micro lens and the camera can use the entire sensor to phase align the image within each dual diode pixel then combine them to capture the image.

As Brant described both the R5 and the 1DXIII utilize versions of dual pixel AF. I also do not know the nuances of those versions to say how their performance would compare in low light. One possible difference is pixel size where the 1DXIII pixels are bigger and I think it is well established that bigger pixels tend to perform better in low light in terms of signal to noise ratio. Not sure if this includes DPAF performance.

In my experience the R5 autofocus system has performed well in various lighting situations. I don't notice any particular difficulty with lens hunting. I have shot many indoor images with relatively low light but usually with a Canon 16-35mm f/2.8 or Canon 24-70mm f/2.8. I have a 35mm f/1.4 but have not yet used it with the R5. Also have had good success with eye detect on humans and animals but I have the SET button programmed to scroll through 3 or 4 different AF settings so I can immediately switch to a zone AF method or spot AF when needed.

I'm sure AF technology will keep moving forward but I can't say the R5 AF is a problem or inferior to DSLR systems .... I don't own a 1DXIII but I do own a 5D MKIV, 1DX and 5DS-R.

[Kayaker72]

Not trying to jump in for Fast Glass, but also to clarify if my assumption was incorrect. I took the question to be a comparison of the PDAF through the viewfinder and DPAF using liveview both on the 1DXIII.

Overall, my key takeaway would be in line with Joel's. I am not having any significant issue with the AF on the R5. The AF definitely performs best with targets within a certain range and with plenty of light, but that is something that is true with all AF systems.

One possible difference is pixel size where the 1DXIII pixels are bigger and I think it is well established that bigger pixels tend to perform better in low light in terms of signal to noise ratio. Not sure if this includes DPAF performance.

I have thought about posting on this topic to see what others think. But, I am not sure this is true anymore. I see it referenced all the time, so I may be incorrect. But, after microlenses were first introduced and optimized, and the architecture size of the sensor chips decreased, I think those steps have largely taken away the pixel size issues. Before light either fell into the pixel well or hit the physical pixel well wall, essentially being lost. For the same side wall thickness, the more pixels you have, the ratio of pixel well area to top of side wall area decreases, so more light is lost bouncing off the top of the sidewalls of the pixel wells. But, microlenses act as funnels directing more light into the pixel well. Then, Canon switched to a smaller architecture, that also decreased the thickness of their sidewalls.

I could easily be wrong. But, I have also played with real world ramifications. What I can say is that I do not see much of a correlation between DR performance and pixel density, but I see a very strong correlation between DR and sensor size.

Another way to look at this, but Quantum Efficiency (efficiency of a sensor at capturing light) of the 5DII was 35%, 5DIII was 47%, and 5DIV was 54%. Move to the most modern sensors, the R6/1DXIII are 55% and 53% respectively (all taken from Photons to Photos website, but based on S/N ratio from DXO Mark). For the record, Sony's A7R III is 59% efficient. So, bottom line is modern sensors are capturing 53-59% of the light, a dramatic improvement over 35% of the 5DII. What is fun, still some headroom to improve into.

But, playing with pixel density, the 5DIV and 80D (same generation of sensor) have the same 54% efficiency. Even go back before that, the 5DSR had 49% efficiency compared to the 47% of the 5DIII. So, almost the same efficiency (5DSR released a year or two later, so perhaps some slight improvements).

[Joel Eade]

Not trying to jump in for Fast Glass, but also to clarify if my assumption was incorrect. I took the question to be a comparison of the PDAF through the viewfinder and DPAF using liveview both on the 1DXIII.

Ah .... that makes more sense to me now. With mirror up in live view the 1DXIII is functioning like a mirrorless camera but instead of contrast detect focusing like many prior cameras used in live view it is using DPAF.

I also wonder if the more powerful battery in 1DXIII has any effect on how fast it can achieve focus in terms of physically driving the lens faster?