Concerns over the Nikon D5 DR Strategy

[shadowblade]

Are there any cameras on the market that reads out 50 million pixels at 50 fps?

At least 30fps - the 5Ds can shoot video at 30fps, and uses every pixel on the sensor to do so. That is, every pixel on the camera is read in order to produce video, even if it's then downsampled to 4k before being written to the card.,

[hjulenissen]

At least 30fps - the 5Ds can shoot video at 30fps, and uses every pixel on the sensor to do so. That is, every pixel on the camera is read in order to produce video, even if it's then downsampled to 4k before being written to the card.,

Sources? I spent 2 minutes with google and could not find an authorative source either way.

H

[shadowblade]

Sources? I spent 2 minutes with google and could not find an authorative source either way.

H

They're not reading every 2nd or 3rd pixel, that's for sure - if they were, they'd produce horrible moire and other artifacts. Canon moved away from the every-few-lines approach to the every-pixel-and-downsample approach after the 5D2.

No idea about sources. I don't look for them when it's both self-evident and consistent with what Canon has done in the past.

[AlterEgo]

They're not reading every 2nd or 3rd pixel, that's for sure - if they were, they'd produce horrible moire and other artifacts. Canon moved away from the every-few-lines approach to the every-pixel-and-downsample approach after the 5D2.

and sources are ?

[shadowblade]

and sources are ?

No idea about sources. I don't go looking for them. You can choose to believe it or not, I don't care either way.

It's consistent with what Canon has done in previous models (apart from the 5D2, the very first with video), makes sense from an engineering point of view and is consistent with the video output from the sensor.

[hjulenissen]

They're not reading every 2nd or 3rd pixel, that's for sure - if they were, they'd produce horrible moire and other artifacts.

Line-skipping was used in highly praised "still image cameras used for video", such as the 5Dmk2. And yes, that camera had lots of artifacts.

No idea about sources. I don't look for them when it's both self-evident and consistent with what Canon has done in the past.

If we are in the "speculation and hand-waving" department, I'd offer mine as well:

The 5Ds is clearly pitched at the stills photographer. Canon will have a video oriented 6D/5Dmk3 replacement soon. I would not expect the 5Ds to be particularly adept (for its price or introduction date) when it comes to video.

So do you have any other examples supporting your claim that "current 42MP and 50MP sensors can sample video at 25 or 50fps, utilising every pixel on the sensor"?

Am I right that the Sony A7rII does line-skipping in FF mode (strange wording in the link below), while the cropped "super 35" (18MP crop of 42MP) is 4k@25fps? Why would the worlds best camera sensor manufacturer do such a thing unless it is somewhat hard to read out all those pixels many times a second? If the camera can actually process and compress 100mbps lossy 4k@25p video, why use anything less than the optimal starting-point (42MP of full sensor area/resolution, properly downsampled to 4k)?
http://www.bhphotovideo.com/explora/video/hands-review/sony-a7rII
"The camera also runs very hot when shooting 4K internally, to the point where overheating could be an issue in certain environments or shooting styles.
...
The full-frame mode uses the whole resolution of the sensor and thus uses pixel binning to down-sample it to 4K. While full frame isn’t quite as sharp as Super 35 and can pick up a small amount of moiré and aliasing in some shots,
"

-h

[shadowblade]

Line-skipping was used in highly praised "still image cameras used for video", such as the 5Dmk2. And yes, that camera had lots of artifacts.If we are in the "speculation and hand-waving" department, I'd offer mine as well:

The 5Ds is clearly pitched at the stills photographer. Canon will have a video oriented 6D/5Dmk3 replacement soon. I would not expect the 5Ds to be particularly adept (for its price or introduction date) when it comes to video.

So do you have any other examples supporting your claim that "current 42MP and 50MP sensors can sample video at 25 or 50fps, utilising every pixel on the sensor"?

Am I right that the Sony A7rII does line-skipping in FF mode (strange wording in the link below), while the cropped "super 35" (18MP crop of 42MP) is 4k@25fps? Why would the worlds best camera sensor manufacturer do such a thing unless it is somewhat hard to read out all those pixels many times a second? If the camera can actually process and compress 100mbps lossy 4k@25p video, why use anything less than the optimal starting-point (42MP of full sensor area/resolution, properly downsampled to 4k)?
http://www.bhphotovideo.com/explora/video/hands-review/sony-a7rII
"The camera also runs very hot when shooting 4K internally, to the point where overheating could be an issue in certain environments or shooting styles.
...
The full-frame mode uses the whole resolution of the sensor and thus uses pixel binning to down-sample it to 4K. While full frame isn’t quite as sharp as Super 35 and can pick up a small amount of moiré and aliasing in some shots,
"

-h

5D2, 7D, 6D and the xxD line up to the 70D use line-skipping video.

5D3, 7D2 and 1Dx (i.e. the newer high-end bodies) use pixel binning.

No idea what the Sony bodies do. I don't use them to shoot video.

[hjulenissen]

5D2, 7D, 6D and the xxD line up to the 70D use line-skipping video.

Sounds reasonable

5D3, 7D2 and 1Dx (i.e. the newer high-end bodies) use pixel binning.

By pixel binning, I assume that you capturing the charge of each individual sensel, combining groups of sensels values into one charge, then reading that value using a single A/D-converter.

Do you have any sources? I can't think of how they would do pixel binning of a Bayer CFA-ed raw signal.

No idea what the Sony bodies do. I don't use them to shoot video.

Since you mentioned 42MP in your original argument, I assumed that you had some opinion on Sony cameras.

-h

[shadowblade]

Sounds reasonableBy pixel binning, I assume that you capturing the charge of each individual sensel, combining groups of sensels values into one charge, then reading that value using a single A/D-converter.

Do you have any sources? I can't think of how they would do pixel binning of a Bayer CFA-ed raw signal.Since you mentioned 42MP in your original argument, I assumed that you had some opinion on Sony cameras.

-h

No, the combination takes place after all the photosites have been individually read. As in, all of them undergo A/D conversion first, then the image is downsampled to the final resolution. I don't know how you could bin the pixels prior to A/D conversion without using a completely different architecture to full-resolution stills.

Don't believe me? Take a single frame from a video sequence of a nonmoving scene, then take a still image of the same subject using the same ISO and shutter speed as was used for video. Downsample the still image to the same resolution as the video frame. You will find that the noise is in the same ballpark. In other words, the video sequence is making use of every pixel on the sensor - if it were only using every second or third pixel, you'd be getting several times the noise.

[BernardLanguillier]

Yet the applications where these improvements will be noticeable are only a tiny proportion of photographs.

If you're shooting fast action in good light, a D4s or 1Dx will also track perfectly (so will a 5D3 or D810). If things are dark but the subject isn't moving so fast towards or away from the camera, or is far away enough that the lens doesn't have to move much, a D4s or 1Dx will also track perfectly. You can't track better than perfectly - the D5's AF won't help where the tracking is already working well.

Another report from real world showing pretty clearly that there is no need to be working on super challenging subjects to see very significant improvements from the AF of the D5:

http://www.dpreview.com/forums/post/57740464

Cheers,
Bernard

[hjulenissen]

No, the combination takes place after all the photosites have been individually read. As in, all of them undergo A/D conversion first, then the image is downsampled to the final resolution. I don't know how you could bin the pixels prior to A/D conversion without using a completely different architecture to full-resolution stills.

Exactly. Then you are not talking about pixel binning, but straight image scaling. Pixel binning occurs in the analog domain to overcome readnoise. Image scaling is (in this context) a sw/hw digital process.

http://www.photometrics.com/resources/learningzone/binning.php
"Binning is the process of combining charge from adjacent pixels in a CCD during readout. This process is performed prior to digitization in the on-chip circuitry of the CCD by specialized control of the serial and parallel registers. The two primary benefits of binning are improved signal-to-noise ratio (SNR) and the ability to increase frame rate, albeit at the expense of reduced spatial resolution."

Don't believe me? Take a single frame from a video sequence of a nonmoving scene, then take a still image of the same subject using the same ISO and shutter speed as was used for video. Downsample the still image to the same resolution as the video frame. You will find that the noise is in the same ballpark. In other words, the video sequence is making use of every pixel on the sensor - if it were only using every second or third pixel, you'd be getting several times the noise.

I don't have a 5Ds. I have a 7D, and I think that we agree that it is doing line-skipping?

Further, I think that your method has some serious limitations. You do know that lossy video codecs throw out something like 99% of the information in an image, using (in effect) complex spatio-temporal transforms? And that removing noise is an efficient pre-processing to help efficient encoding? Comparing noise levels visually in such cases is very hard.

I would rather suggest you (if you have one) record video of some still resolution chart. If there is line-dropping going on, you should see clear examples of spatial aliasing, rather than the (generally preferreable, IMO) smooth reduction in high-frequency reproduction.

-h

[BernardLanguillier]

First AF comparison on moving subjects with the 1DXII at Dpreview:

http://www.dpreview.com/news/6990762465/motor-drive-and-motocross-with-the-nikon-d5-and-canon-1d-x-ii/3

Sounds like the D5 wins on that metric.

It is indeed so good AFwise that I would have been amazed if the 1DXII has been even better.

Cheers,
Bernard

[dwswager]

First AF comparison on moving subjects with the 1DXII at Dpreview:

http://www.dpreview.com/news/6990762465/motor-drive-and-motocross-with-the-nikon-d5-and-canon-1d-x-ii/3

Sounds like the D5 wins on that metric.

Cheers,
Bernard

Based on this, the 7DmkII doe not compete well with the 1DX, 1DXmkII or D5.  I'd love to see how the D500 fares against the D5 in focus speed, lock on and tracking. 

For sports, there are times when pure burst rates can make the difference, but also lots of times when it doesn't (the action moves too fast for even a 10-14fps camera to catch the peak), especially when light is low and shutter speed retarded. 

[BernardLanguillier]

A few random images captured with the D5 recently.







Don't know why, but the camera seems to work.

Cheers,
Bernard

[Bart_van_der_Wolf]

Don't know why, but the camera seems to work.

Then maybe it is the photographer who can make a difference, since it couldn't be just the camera ...

Cheers,
Bart

[BernardLanguillier]

Then maybe it is the photographer who can make a difference, since it couldn't be just the camera ...

Ah yes, maybe.

Now some cameras are better than others at not introducing hurddles btwn a photographic intent and its realization. Funny to say this considering the bulk of the beast, but the D5 does a great job at removing itself from the story.

Cheers,
Bernard