Re: Sony A7RII - FIRST video sample!!!

DLD said:
Canon still beats Sony on glass
Click to expand...

I'm not sure about that.

As a photog, I'd be terribly happy with the two Batis and a 55mm f/1.8. Much more than with three Canon primes adding up to the same cost, though on Canon you can get the Sigmas and the 24mm and 50mm Art, plus the Canon 85mm f/1.2, would make a great set too.

And as for video, being able to mount nearly any manual lens on your camera really is a great advantage.

What I get is that people already have good Canon glass, and it's expensive to switch a full lens set (and using Canon glass on a7 bodies kind of kills the size advantage of the mirrorless system).
 
I agree with you on sharpness, Samuel, but I guess he's speaking about range (focal length, fast zoom lenses, etc), isn't it? ; )

E :-)
 
Well, if you need a fast zoom, it's going to be big, so going mirrorless doesn't help you a lot. But *I* shoot primes, and the FE mount already has a very nice range for that (a 135mm f/2.0 would be the only one I'd miss). As usual, it's a very personal thing, it depends on what type of lenses you like to use.
 
It lacks fast zooms and fast teles, as for instance. It beats on the lightweight department, though.

Those true Zeiss (whether people say they are not whether people are educated on the way a brand value is taken ;) are priceless, anyway.
 
The s35 mode on the a7rII most certainly is the same crop as APS-C (rounded in video mode for read-out/oversample convenience). I don't even need to see the camera to claim this. Anything else would be stupid.
 
cpc said:
The s35 mode on the a7rII most certainly is the same crop as APS-C (rounded in video mode for read-out/oversample convenience). I don't even need to see the camera to claim this. Anything else would be stupid.
Click to expand...
But they had announced they had found some mojo tech to workaround the issue such as a faster readout speed, I guess.

In any case, there's also the fact Dan Chung actually tested with a tele zoom lens in crop mode. How this can be accepted as worthy to test its rolling shutter performance is something to also take in consideration.
 
For stills the crop is 1.54x (7952/5168).

Usually the crop in video mode is exactly the same as in stills.

Somebody speculated they could have gone for 1.38x (7952/5760) to get a simpler downsampling to QHD by reading 3x1920. But the diagonal for 3:2 with 1.52x crop is 28.47mm, with 16:9 1.38x it is 29.93mm, that's enough of a diference that some lenses will show a hard crop.
And also: shooting 16:9 stills in APS-C mode it gives you 5168x2912 files, which is exactly the 15Mpix oversample that the Sony rep cited in that lenghy interview about the new sensors.
I think we can almost close this one: an actual test would seal it, but it must be 1.54x, just as in the APS-C stills mode.
 
So, is it fair to test rolling shutter with a 110-310mm FF 35mm equivalent focal length? I am still waiting for such information from the shooter who looks like to unknow. Who would use a tele to track any motion in-frame?

A more realistic approach to a real shooting case study is mandatory if we actually want to evaluate the performance of this device under real circumstances.
 
My measure works no matter what lens you use. It just demands that the camers is on a tripod and there's a vertical line going through the whole frame, from top to bottom.

That way I get a measurement that is in ms, and will not change no matter what lens you use. Use a longer lens and with the same motion you'll get more skew, use a shroter lens and with the same motion you'll have less skew, in ms the result will be the same (the inter-frame displacement changes proportionally to the focal length, and the final measure in ms is invariant).

So yes, I think it's perfectly valid.
 
Samuel H said:
My measure works no matter what lens you use. It just demands that the camers is on a tripod and there's a vertical line going through the whole frame, from top to bottom.

That way I get a measurement that is in ms, and will not change no matter what lens you use. Use a longer lens and with the same motion you'll get more skew, use a shroter lens and with the same motion you'll have less skew, in ms the result will be the same (the inter-frame displacement changes proportionally to the focal length, and the final measure in ms is invariant).

So yes, I think it's perfectly valid.
Click to expand...
Samuel, I have nothing to object about your measurement method. Thanks for your clarification, though.

I am referring the perceived skew instead. A teleobjective focal length will obviously give you a worse perception on the skew. And the outcome will seem as horrible as actually already is from your numbers :-D
 
The point is how to minimize that or even try to surpass it. Pending on the shooting style. 5-axis? And so on.

I'd rather have seen much more in order to not say the jury's still out. In a substantial part, at least :-(
 
ectobuilder said:
I was looking into the Loxia's too and I was very dissapointed with the close focusing distance. This is a weak point for most E-Mount lenses, but as a high end prime I expected more. To me, an elite prime should have a close focusing distance, it is part of the versatility of a prime that I've come to expect.

Here is a list of lenses with the closest focusing distances calculated as the closest distance from the subject to the front of the lens. Normally the "closest focusing distance" is calculated from the image plane, but this is misleading if the lens has a shorter length, which most E-Mount (FE & SEL) lenses are. So when you take this into account, the E-Mount lenses perform poorly, with the exception of a few. Most that do well aren't made by Sony or Zeiss, it is Samsyang/Rokinon/Bower.
Click to expand...

I was all Canon many years ago and sold everything when i moved to Sony.

I find my lens choices on Emount vastly superior to Canon because of the short flange distance.

I use everything from Zeiss ZA Zooms, Russian Lomos, Contax Zeiss and Range finder glass including pretty much all the voigtlander lenses. IMHO those voigtlanders are exceptional. As for close focus, i have a number of helicoid adapters (leica to emount) which move the close focus even closer, my 12mm voightlander will focus a few cm's in front of the lens. Then for normal rangefinder i can bring the focus distance to the normal SLR types.

One thing i don't have are very many native e mounts and i would agree they're a bit blergh. But as a mount system i don't think there is anything comparable anywhere else.

cheers
Paul
 
Samuel H said:
* In FF mode it's probably skipping lines and columns, and reading exactly 3840x2160 photosites.
Click to expand...
At the moment, I'm still sticking to my theory that in FF mode it reads exactly twice that number horizontally and vertically (then pixel bins on a 2x2 basis). To do as you say, (reading exactly 3840x2160 photosites) then the question would be exactly how it was doing it?

It couldn't just read a row, skip a row (same for columns) as that would mean it would only ever be reading green and red photosites, and never rows with blue in (or vice versa). Read two, skip two, then deBayer ignoring the gaps would get over that - but I think the results would be far from good, and the quality drop cf the s35 mode very evident, even without a chart.

And reading Dan Chungs report he says "The second big surprise is that the full-frame 4K mode also looks remarkably good." and "The 4K full-frame output of the a7R II is really excellent." It's hard to reconcile those statements with any crude form of pixel binning - which is why 2x2 binning of 7680x4320 photosites still seems most likely to me?

Wishful thinking? Maybe, but let's wait until we get some accurate figures for horizontal crop figures relative to FF stills mode. Those will show definitively how many photosites are being used for each mode.
 
harddrive said:
At the moment, I'm still sticking to my theory that in FF mode it reads exactly twice that number horizontally and vertically (then pixel bins on a 2x2 basis). To do as you say, (reading exactly 3840x2160 photosites) then the question would be exactly how it was doing it?

It couldn't just read a row, skip a row (same for columns) as that would mean it would only ever be reading green and red photosites, and never rows with blue in (or vice versa). Read two, skip two, then deBayer ignoring the gaps would get over that - but I think the results would be far from good, and the quality drop cf the s35 mode very evident, even without a chart.

And reading Dan Chungs report he says "The second big surprise is that the full-frame 4K mode also looks remarkably good." and "The 4K full-frame output of the a7R II is really excellent." It's hard to reconcile those statements with any crude form of pixel binning - which is why 2x2 binning of 7680x4320 photosites still seems most likely to me?

Wishful thinking? Maybe, but let's wait until we get some accurate figures for horizontal crop figures relative to FF stills mode. Those will show definitively how many photosites are being used for each mode.
Click to expand...

Sony says "binning" not skipping.

If the low light is within a half stop of the A7S this is a MONSTER.

If not it's still pretty sick.
 
harddrive said:
At the moment...
Click to expand...

* What many sensors do is read two, skip two. For example, the CMV12000 that is inside the BM 4K cameras definitely has that option. I believe many Canon DSLRs were doing "read one, skip two", which is why they get those horrible rainbows all around.

* The a7R II has such small photosites that it will create very little moire with a "read two, skip two". Basically no color moire (because there are always red green and blue samples very close to each other), and aliasing only if you use very sharp glass.

* The "skip columns too" part is what I'm not sure any cameras are doing. From 5D3 RAW video samples, I think it's doing "skip rows, but bin columns", and that seems to be enough to kill aliasing and moire almost completely, that's why I think other cameras are discarding columns too (they probably read them, but discard the data because it's faster than digital binning).

* Given that on the a7R II RS is slower in APS-C than in FF mode, it can't be doing digital binning. It either does analogue binning (which is absolutely awesome but requires extra circuitry in the sensor), or does line skip (which is usually implemented precisely because you want to have faster rolling shutter). Being BSI gives the sensor in the a7R II some chance that it may actually do analogue binning, but I'd still think it's a small chance, it's kind of a big deal and I think we would have heard about it (with binning you get no aliasing/moire, it's skipping that causes that; yes, I'm puzzled about why Sony talks about binning and not skipping, it makes no sense to me).
 
Binning actually doesn't kill aliasing. All on-sensor binning is unweighted sumation of all the pixels in the bin and the resulting binned pixels are never on the ideal interpolated position in the grid. There is some relatively recent research that suggests methods specifically targeted at debayering binned mosaics; these methods do fight resolution loss and decrease aliasing, but I don't know if any new camera actually implements something like this. On the 5dm3 I occasionally need to treat aliasing in post (nothing serious though, and moire is under control).

Software bayer data binning is much, much better since it can do various perfectly weighted interpolations, but it will still manifest some aliasing even with alias attenuation filters. It does require full sensor readout of course.

An aliasing test of the FF mode of the A7rII will show what's happening. I wouldn't hold my breath for proper binning though.
 
Samuel H said:
* What many sensors do is read two, skip two. {lines}
Click to expand...
Agreed.

Samuel H said:
* The "skip columns too" part is what I'm not sure any cameras are doing.
Click to expand...
They have to read all the columns within a line - but may then just not use two out of every four read out within a line. The clue is if zone plate results are symmetrical. If that's true, the camera *MUST* be doing the same for rows as columns - if it wasn't, the zone plate results wouldn't be symmetrical. The approach that Panasonic use with the GH series and the AF100 is (as you say) "read two, skip two" for lines, but they must then discard half the samples per line - otherwise the results wouldn't be symmetrical. (And they are; see http://www.provideocoalition.com/ag-af100_and_pmw-f3_on_the_charts/page-2 .)

From other features of the chart, and knowing basic characteristics of the sensor (pixel count etc), the only system that makes sense to give those results is reading a 2x2 block, ignoring all the other photosites in a 4x4 block, and then doing a direct R,G,B read of each 2x2 block - no deBayering. This is also exactly what the FS700 does in high frame rate modes. The pattern is exactly the same as for the Panasonic cameras, it's just the figure for resolved lines that differs. It's a very valid technique, and has a unique "fingerprint" on charts.

Similar is to take the entire 4x4 block and bin together all the corresponding R,G, and B sites. It won't change the basic resolution - but will reduce the aliasing (imagine the cancelling effect of fine white-black lines over the entire 4x4 block), and should improve the sensitivity.

Next step is to simply take ALL the rows, all the columns, and again directly read in 2x2 Bayer blocks. Exactly as the Canon C series cameras do. In a way it's stretching it to call it "binning" as in each block only two G sites get averaged. For R and B the corresponding output is read directly from the photosite value. But it's certainly not "skipping" any more. Which would be why Sony talks about binning and not skipping

And the point is that this works extremely well when the number of blocks corresponds to the output resolution. For the Canons that's 1920x1080 blocks - I believe for the A7R II it's 3840x2160 blocks

Samuel H said:
* From 5D3 RAW video samples, I think it's doing "skip rows, but bin columns", ......
Click to expand...

But for obvious reasons you can only bin photosite outputs of the same colour. So since for red and blue you only get 1 photosite in 4 of that colour, it makes the geometry of exactly which to bin together quite difficult, especially if combined with line skipping. Assume binning is happening in square blocks of 2x2, 3x3, or 4x4 at a time (to give three binned outputs for each colour, for each block) and the geometry makes sense.

So start with 7680x4320 and bin 2x2 to get 4K, bin 4x4 to get HD.

Samuel H said:
* It either does analogue binning (which is absolutely awesome but requires extra circuitry in the sensor), or does line skip (which is usually implemented precisely because you want to have faster rolling shutter). ....... I'd still think it's a small chance, it's kind of a big deal and I think we would have heard about it (with binning you get no aliasing/moire, it's skipping that causes that; yes, I'm puzzled about why Sony talks about binning and not skipping, it makes no sense to me).
Click to expand...
And my money is still on analogue binning, but it does indeed need proper tests - and by that I mean charts. And as said before, it will be possible to draw a lot of conclusions from horizontal crop comparisons in the various modes.
 
Yet as far as roliing shutter goes, about 30ms (a7S) + right focal length + careful pans are able to produce this outcome:

 
You must log in or register to reply here.
Back
Top