Here is an off the wall thought. Is there any chance that a 5DMkII class of sensor could eventually be set up for 4K video? This might lessen the compromise between sharpness and moire.

Ideally, we'd be able to read out the whole sensor for each frame, which is what RED does. Unfortunately, the 5D MkII sensor/system isn't fast enough to do this. Even when skipping every third line, it takes 75% of a frame interval to read the data.

So, yeah, it could read out all the data, but no faster than 13.3 frames per second. Rolling shutter artifacts at that rate would make this a flubber cam!

alias
 

All:

I have spent an hour or two looking over the messages on this thread and another dvinfo alias discussion on this camera. My conclusion is that by subsampling vertically on the sensor pixels, rather than averaging over them, to get down to 1080 lines, the camera designer has given up the main line of defense against the aliasing that has been sometimes seen. While it is true that defocusing the lens and/or adding in an optical anti-aliasing filter will help alleviate the problem, unfortunately there is no real substitute for the above mentioned decision to sample the sensor output without a vertical average first. As some of you have mentioned, this may well have been done because the hardware read speed (sensor+computer) would not permit the vertical average together with a video frame rate output.

John Woods

Actually, the optical solution can totally remove the aliasing, but because we can't get a brickwall optical filter, you will lose some detail. In theory, we can get the same results as any camera with a native 1080 vertical line sensor.

It would have been awesome had they been able to sample the whole sensor though. Not only could we get high resolution without aliasing (given a powerful digital filter), we could have gotten three times the light sensitivity(!)

fill-factor
 

You forget about the fill-factor. For this camera, in video mode, it just ins't there. So the vertical sampling is much harder to account for with a practical anti-alias filter.

John

Tom, thank you for taking the time to measure and post you data.

Numbers do matter.

I'm going into shoot day 10 participating in a feature length movie shot on the 5DMII. This piece will show on a theater screen, so I'm very curious as to how it will hold up when projected.

Yes, we have run head first into the moire problem, it was extremely prominent one of our shots.

There are many important criteria for measuring a camera. Low light performance is a great 5DMII attribute. The pretty, painterly, color representation is considered pleasant to many, especially those in advertising.

I consider overall accuracy in rendering a frame the most important factor when evaluating any camera. Many cameras these days produce 'pleasant, pretty' images but not accurate representations of what an eye would see. I think accuracy is the hardest to engineer for and a good meter by which to evaluate a camera.

Tom: what picture style did you have your 5D2 set to when you made that capture? In my experience from this past weekend, When sharpening was up a few notches from the lowest settings, the image looked very nasty, with intense halo around contrasting edges, sort of like the result you got..

It was "standard" style. The sharpening was the default '3' setting. In retrospect, I would have shot with "neutral" to more closely match the other cams.

IMHO, if you're shooting with a default preset, neutral with contrast, sharpening and chroma dialed all the way down is the best bet for a neutral image you can fiddle around with in post. OTOH, I disagree with all the artifact discussion. Yes, it has a rather naty tendency to moiré and sometimes fine horizontal detail in focus can look pretty nasty, but nothing like the artifacts in cameras like the XLH1 or XHA1. First you have the very noticeable resolution drop when going for a progressive frame mode. Next, you need a good uprezzing software to avoid nasty stairsteping from going up from HDV's recorded 1440 to 1920. And last but not least, the absolute worst artifact in HDV cameras is the horrible horrible chroma subsampling, which, along with the resolution subsampling, not to mention the awful chromatic abberrations the stock lenses produce, makes color edges look like atari games from the 1980's. Pulling decent chroma keys in HDV requires some serious tools and patience. The 5D's image is world's ahead of these cameras just because of those three things, not to mention the infinitely better color fidelity, DOF, clean abberation free lenses, really wide wide angles, tilt-shift lenses, picture styles and higher bit rate to mention but a few.

My .02

The 5D2 has a lot to recommend as you say, however it's problem with moire and aliasing is worse than either of the two HDV cameras you mentioned.

Update:

I just did a quick video to test for the effectiveness of turning down the sharpness by choosing 'neutral' style and sharpness '0'. It has no effect on the aliasing, just makes the picture too soft but still with visible artifacts.

Out-of-focus optical blur is highly effective if you are able to use depth of field control to blur out the distracting parts.

Meh. Most the footage/stills look good to me. Beyond calibration, I stay away from numbers.

Lumpy edges?
 

These are great comparisons. Every time I think of the wonderful wide angle shots, the wafer thin DOF effects and the incredible low light performance, I'll refer back to shots like these which demonstrate lumpy diagonals, and the moire. And the fixed exposure range that can't be pushed/pulled further than what you get.

I'll stick to the plan of a second EX1.