Nikon D800 with video features

[Sareesh Sudhakaran]

Originally Posted by David Heath

The main reason DSLRs suffer from moire is that they (so far) all have to ignore a percentage of the photosites to achieve a fast enough frame rate - hence the averaging is much less than ideal, hence moire. Initially they would skip several lines at a time (really bad), more recently they directly read in 2x2 blocks (as the C300) but (unlike the C300!) skip every other block horizontally and vertically.

A question: Does this mean the OLPF won't make much of a difference as far as video is concerned?

[Konstantin Kovalev]

A question: Does this mean the OLPF won't make much of a difference as far as video is concerned?

An OLPF has to be just strong enough to allow the camera to resolve detail up to it's maximum resolution to avoid artifacts. With or without the filter, the camera is inherently designed to capture 36mp, compared to HD which is a mere 2MP.

The only way a camera with such high resolution could deliver an artifact free downsize, is if it read pixel data from every photosite, as David wrote in his post above. I highly doubt there are any cost-effective processors out there that could handle a 36mp signal @ 30fps and down-sample it to HD in real-time, using a quality interpolation method.

[Josh Dahlberg]

Originally Posted by John Vincent

The moire issue is - as usual - way overblown. Moire might be the biggest non-issue in the camera world - "normal" people could care less. Don't shoot a movie about brick buildings or stripped shirts, and you're golden.

You're absolutely right about "normal" people John, but for me shooting DSLRs moire is *the* issue, along with its buddy aliasing.

The GH2 is the only DSLR I've used to date that adequately suppresses moire and aliasing, and produces a detailed image. I'm waiting for a FF camera that can offer the same detail with good control of artifacts.

There's crazy moire at 3.30 in the music vid... I'm hoping this is something to do with Youtube... but it looks disturbingly like 5DII style moire to me - compression doesn't do this does it?

[David Heath]

Originally Posted by Sareesh Sudhakaran

A question: Does this mean the OLPF won't make much of a difference as far as video is concerned?

Let's be clear, I don't know what the characteristics of this filter are, so can only speak in generalities. But as said before: "As far as the OLPF goes, then if that was optimised for video, the camera would be effectively hopeless as a stills camera ".

By that, "optimised for video" implies reducing the definition to something suitable for 1920x1080. Great for video - but your expensive 36 megapixel DSLR then starts to simulate a 2-3 megapixel in stills mode!

Originally Posted by Konstantin Kovalev

The only way a camera with such high resolution could deliver an artifact free downsize, is if it read pixel data from every photosite, as David wrote in his post above. I highly doubt there are any cost-effective processors out there that could handle a 36mp signal @ 30fps and down-sample it to HD in real-time, using a quality interpolation method.

Yes, but that's assuming reading the sensor, doing a full deBayer (for 36 megapixels!), then a quality downconversion - at 30/60 times a second. But, the other possibility is SIMPLY and DIRECTLY reading the sensor to derive the 1080 signal directly with does away with any need for deBayering and down or up conversion. This is exactly what the C300 does - "problem" is that a 8 megapixel sensor is nowadays considered inadequate for high quality stills. Since the C300 is a video camera, period, it's a non-issue, but a completely different story for a DSLR!

Hence the thought of the 33.1 megapixel 16:9 sensor. Yes, all the photosites need to be read every frame, but it should be possible to use binning techniques to reduce that workload without compromise. Most importantly, no deBayering or up/downconversion are necessary to get a 1080 signal.

[David Heath]

Originally Posted by John Vincent

The moire issue is - as usual - way overblown. Moire might be the biggest non-issue in the camera world - "normal" people could care less. Don't shoot a movie about brick buildings or stripped shirts, and you're golden.

Originally Posted by Josh Dahlberg

There's crazy moire at 3.30 in the music vid... I'm hoping this is something to do with Youtube... but it looks disturbingly like 5DII style moire to me - compression doesn't do this does it?

Ah that it was so simple as avoiding striped shirts etc! :) Josh has the key with the comment about compression.

I heard a great analogy recently whereby moire was compared to food going off (!) Leave some food past it's sell by date and you may get two effects. First is a change in appearance/taste/smell - the analogy here is moire you can easily see with the eye. Second is growth of something which will give you food poisoning - but may not be obvious when you eat it.

And it's the same principle with moire. You can get artifacts which are hardly visible on camera footage - but can play havoc with compression systems further down the chain. Hence footage can look great straight off the camera - but degrade far worse than expected down the production chain. Exactly as some bad food may seem OK to eat - but make you seriously ill a few hours later.

That's why "normal" people DO need to care about moire and food hygiene - even if they don't seem a problem at first viewing or eating!

(If all that seems unlikely, then the characteristic of moire which can cause many problems is that it "ripples backward" in the opposite direction to the object it's associated with. That can confuse coders at low bitrates which use motion analysis - an object can seem to be moving two ways at the same time - and waste bitrate in the coding. Put simply, the end result has the appearance of what it would have been if no moire and a lower bitrate used for the encode.)

[Tim Polster]

Great insight David. Why it pays to go a bit deeper and know the tech behind stuff in this day and age.

[Jacques Mersereau]

This camera is exactly what a ton of people hoped Canon would do and is exactly WHY I am so disappointed with Canon.

Okay, so the D800 isn't perfect - but for $3K? Compared to the Canon C300 for $16K?!!!

Come on folks, what I saw on those clips looked darn good to me.

AND, though this is far from a done deal, there will be tons of hackers and programmers getting inside
and *possibly* having some cool D800 mods. The camera has USB 3 out. If enabled, we could be recording RAW at 12bit or maybe 10bit S log. Who knows? The thing is, at $3K, that means for around $10K I can put together a whole kits that will do just fine.

It really sticks in my crawl that Canon, with what it has done with both the price points of both the C300 and DSLRs, has FORCED me to move to Nikon. That is not what I wanted, but there it is.

Congratulations Nikon. Not to understate it, but you just handed Canon its *$$.

[Don Parrish]

The ball is definately on Canon's side of the net,


Well Canon ??????



P.S. C300 footage looked much better.

[John Vincent]

"P.S. C300 footage looked much better. "

Jeez, it better - it costs more then 5 times as much.

[David Heath]

Originally Posted by Mikael Couderc

Well, Dpreview mentions that when in video mode, the D800 only uses about 91% of the sensor width (hence slightly changing the FOV). One has to wonder if that's not to make the number of pixels a more manageable total to allow a better downsampling to HD video.
I mean, just speculating here, but why 91%?

I've not been able to see that reference - could you point me to it?

As to "why 91%", then I suspect if so it is actually 91.3%. :-) No, not a joke, it's just that 91.3% will be 6720 - and that is 3.5x 1920. It's hard to believe that's a coincidence, and may be a strong pointer to the internal workings in video mode. It would infer 960x540 7x7 blocks of photosites, and various ways of dealing with them could be worked which would all give direct 1920x1080 output - no up or down conversion.

Theoretically, that should give pretty good results. Not as good as my hypothetical 7680x4320 chip, and certainly not as good as the C300 (for video!) - but possibly better than much other video from "designed for still" sensors.

Regarding moire/aliasing etc, then after what I said previously about the REAL problems with moire being it's ability to screw up compression, then it's worth mentioning again that it's the MOTION of the aliases that really causes the problems, the way they move counter to the real objects. For STILL images this is obviously nowhere near as big a problem, any aliases may just make edges look a little "busy".

[Josh Dahlberg]

Originally Posted by Jacques Mersereau

This camera is exactly what a ton of people hoped Canon would do and is exactly WHY I am so disappointed with Canon.

Okay, so the D800 isn't perfect - but for $3K? Compared to the Canon C300 for $16K?!!!

Come on folks, what I saw on those clips looked darn good to me.

To me those clips don't surpass 5DII quality - would anyone be surprised if a 5DII had shot those? On paper the D800 is great, but image-wise it appears like Nikon is playing catch up.

The clips benefit from high production values, but the video quality itself looks soft with fairly clear moire/aliasing issues. The C300 is priced at a premium, but for that you get detailed, artifact free images. The D800 images (to my eye) don't convince, but I'm really hoping this is Vimeo/Youtube at fault because I can't afford a C300!

Canon is about to bring the 1DX to market (annoyingly without a headphone jack and clean HDMI). The 5DIII is surely just around the corner. They've had more than 3 years to come up with serious improvements to their original DSLR video implementation and they have promised (in the literature) much improved suppression of rolling shutter, moire and aliasing, along with superior internal codec; I'm quite optimistic the new generations of Canon DSLRs will handily beat the image quality we're seeing here.

[Josh Dahlberg]

Originally Posted by David Heath

I've not been able to see that reference - could you point me to it?

Nikon D800 Preview: 3. Key Technology: Digital Photography Review
"Movies can also be shot at two different crops from the sensor, FX and DX. This makes it easy to vary the field-of-view for grabbing footage, even if you’ve got a prime lens mounted. However, the ‘FX’ size is a significantly cropped version of the full sensor (it’s 91% of the sensor’s width), so the field-of-view will be a little narrower than you’d expect for any given focal length."

However, DPreview also reported in its D4 preview that the 16mp camera also employs a 91% crop in FX video mode, so it's not sensor specific...

[Wacharapong Chiowanich]

In terms of potential for better image quality in video, the D4 then seems to be superior as it has a more sensitive sensor (with similar technology), far less pixel count, thus much less raw data for the IP (similar or identical) to deal with.

Whatever the tech behind the D800's video is, I can see no reason why Nikon can't do better with the D4 for about twice the price.

[Sareesh Sudhakaran]

Originally Posted by Josh Dahlberg

To me those clips don't surpass 5DII quality - would anyone be surprised if a 5DII had shot those? On paper the D800 is great, but image-wise it appears like Nikon is playing catch up.

The uncompressed 8-bit 422 is the big difference, isn't it? We'll know soon enough whether its worth anything in a few months.

[Sareesh Sudhakaran]

Won't the D800 in DX mode equate to the D4 in terms of video quality? Wouldn't the sensor readout be approximately the same, if the technologies were developed together?

Is it safe to presume that the D800 will be 'sharper' in DX mode?