The Downside to More Pixels

(commentary)

We had two missiles fired in the “massive pixel” realm this week, the 50mp Canon 5DS and the 63mp Olympus E-M5II. Boy those are missiles with big numbers, but what’s behind the covers?

In Canon’s case, they’re already getting a lot of pushback on dynamic range. Canon has for some time now been lagging the Sony/Nikon gains in this realm, and the new 50mp cameras don’t change anything in that respect, it seems. Even Canon, when directly asked, is saying that the new camera won’t gain dynamic range over previous ones. 

That’s not a bad thing, per se. But it does mean that the pixel push is really only about sampling density, not improvement in other aspects of the sensor. Even the Canon faithful seem to be a little disappointed by that, though they’ll take any improvement they can get, obviously. The 5DS should be a great studio camera, for example, as you control the dynamic range of a scene with lighting. For landscape work, those Canon shooters will likely still want to look at HDR to push their work upwards more towards medium format results, though.

Meanwhile, over at Olympus, the E-M5II takes a massive 128 megapixels worth of data for their new Hi-Res mode. Yet it outputs that as only half that in raw (63mp) and one-third in JPEG (40mp). Looking closer, it appears that the E-M5II doesn’t actually take eight perfectly-exposed shots, nor does the raw file give you access to the original data for those shots. By perfectly exposed, I mean 8x the metered shutter speed should be the length of time the eight shots take, all at the metered shutter speed. But they don’t. They take far less. So the camera is underexposing the shot stack and counting on the alignment of the R, G, and B patterns to make up for the noise that would be produced in doing so. 

When I first saw the 40mp JPEG samples from the camera, I wondered about them. It seemed they were a little low in contrast to me, with a low noise threshold. Yet also high in acuity. That’s an odd juxtaposition. What we know at this point is that for raw files, Olympus does some sort of interpolation of pixel data from the aligned and half-aligned four-shot sets. Worse, the red and blue data is clipped (at 2000 max in 12-bit data terms after black subtraction, while the green channel is clipped at 3840). If we were really using all 12 bits, we should have as many as 4096 values at each location, though this gets us in a discussion about what “black” is. 

What seems to be happening is the multiple shots are lowering noise, lowering dynamic range, and increasing resolution/acuity. Oh, and for some reason the file format is packed, but not compressed, meaning that file sizes are breaking 100MB (and remember, a lot of that data is clipped, wasting bits that aren’t getting compressed). 

Olympus users will ultimately be a tiny bit disappointed in these new big megapixel files, I think. Especially since there seems to be a great deal of confusion about how data gets into those Olympus raw files in the first place. We have some sort of “cooked” raw, not the raw data from the original eight exposures, thus we’re left at Olympus’ whim. 

When I saw what they were doing, I immediately asked myself why couldn’t they provide us with either (1) a full 32mp 12-bit RGB file (essentially a really clean 32mp TIFF), or (2) the eight data sets so we could combine them as we see fit? 

Olympus missed out on creating the best possible black and white digital camera we’ve seen, by the way. We should have a minimum of 32mp really clean luminance resolution, but with full color information that can be used to “filter” the monochrome results without the usual noise propagation. 

Curiously, the weaknesses of both the Canon and Olympus efforts leaves the door open for Nikon/Sony to provide both higher resolution and higher (than Canoly) dynamic range. As usual, there are no winners in the tech spec race, only temporary leaders. For the moment the leaders (outside of medium format) are:

  • Dynamic range: Nikon/Sony
  • Direct pixel count: Canon
  • Per pixel noise: Sony; Nikon close


A week ago it would have been Nikon/Sony in all three categories. So progress?

© Thom Hogan 2015 / All Rights Reserved bythom.com  @bythom #bythom