Did anyone notice that 5D3 native image resolution is exactly 1920*3 x 1280*3 (5760x3840). It is 3x3 pixel binning for video.
I've been wondering why no one has mentioned this when discussing Canon's supposedly odd 22.3MP resolution choice. This also seems to be a much overlooked factor when assessing moire and rolling shutter improvements.
The resolution certainly seems like a strange choice; it must have to do with video....
However, 1920X1080 bayer pattern (3x3 pixel binning) isn't that sharp. The F3 has a 3+ megapixel sensor and the Alexa a 4+ megapixel sensor to achieve a sharp 1080p image. In theory, bayer sensors can resolve about 70% of their stated resolution (not even taking aliasing into account)--so you need almost four megapixels to get a really crisp 1080p image. And this puts the 5DIII's 1080p mode at closer to 720p, about what the previous cameras have resolved. So that's disappointing.
The latitude, decrease in moire, decrease in skew (fingers crossed), and great looking colors still count for something, though.
So... disappointed because it's not up to par with the Alexa? I would say while it's nothing to gush at we shouldn't be disappointed because it's not competing with cameras that are about 10x the price.
It's not just the Alexa...virtually every decent video-specific 1080p capable camcorder records at a higher resolution than bayer 1080p and scales down--because bayer sensors aren't that sharp at a per-pixel level. Even models in the few hundred dollar range oversample. It just seems strange that the sensor would be designed around video resolution, when the video resolution it's designed around produces a fuzzy image. A weird compromise in general, though if the recording were done with 2:1 binning and then scaled down from there...that could look incredible.
And seriously, don't start the "not as good as x" debate. The Alexa may have better resolution, but it also has way less skew than any other CMOS-based camera, a great codec, more latitude than anything else, and awesome highlight rendition with a smooth bloomy look. Even the c300 isn't as good. DSLRs aren't close. Meanwhile, the fs100 (which uses the f3 sensor) is only 33% more expensive than the 5DIII and has a properly implemented super35 bayer sensor. But it's a dedicated video camera, so again a somewhat unfair comparison.
Then again, the 5DII apparently debayers from 5616x1053. Which means the 5DIII will have LESS resolution. Overall, the 5DIII should have 30% LESS resolution than the 5DII if this pixel binning stuff is true.
I get that the 5DIII isn't a video camera. It's just odd that the sensor's resolution was apparently chosen to improve video quality when a multiple of 1080p shouldn't help things one bit. I suppose maybe this is a limitation of the digic5 chip--it can't debayer, downscale, and compress that fast...who knows. Certainly the 5DIII video samples have less aliasing than the 1DX samples, which is a great start.
And that said...a reduction in aliasing is a huge and substantive improvement relative to the introduction of 1080/60p, so I applaud Canon for that (the Alexa and Red didn't do this to start because it's no one's priority--you all really use slow motion that much?! rent a phantom already....). This might still be a nice camera for video. Looks amazing for stills.