Continuous bracketing on the alpha 7 cameras

Started Feb 13, 2015 | Discussions
JimKasson
JimKasson Forum Pro • Posts: 21,718
Continuous bracketing on the alpha 7 cameras
4

Automatic bracketing is implemented on the Sony alpha 7 cameras as a shutter, or drive, mode,. There are two modes, continuous and single shot. If you use continuous mode, the raw (pre-compression) bit depth is changed from 13 bits to 12 bits, just like when the camera operates in continuous shutter mode without auto-bracketing.

If you're doing handheld HDR, you want continuous auto-bracketing. If you use single shot auto-bracketing, there will be greater camera motion between shots, more work for the auto-registration feature of your HDR program to do, and less chance that it will do the job precisely.

But with the a7 cameras inflicting the bit-depth tax when you use continuous auto-bracketing, you're going to lose about a stop of DR just for invoking that mode. So, if you select as a bracketing sequence 0, +0.3, -0.3 stops, you're actually going to be worse off than if you'd not used auto-bracketing in the first place.

To my way of thinking, the minimum continuous auto-bracketing span that makes any sense is 2 stops, and even that is less than a one-stop improvement over no auto-bracketing at all. So use 0, -1, +1 or more when you use continuous auto-bracketing.

-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
Sony a7 Sony a7 II Sony a7R Sony a7S
If you believe there are incorrect tags, please send us this post using our feedback form.
Camley Senior Member • Posts: 1,763
Re: Continuous bracketing on the alpha 7 cameras
1

JimKasson wrote:

Automatic bracketing is implemented on the Sony alpha 7 cameras as a shutter, or drive, mode,. There are two modes, continuous and single shot. If you use continuous mode, the raw (pre-compression) bit depth is changed from 13 bits to 12 bits, just like when the camera operates in continuous shutter mode without auto-bracketing.

If you're doing handheld HDR, you want continuous auto-bracketing. If you use single shot auto-bracketing, there will be greater camera motion between shots, more work for the auto-registration feature of your HDR program to do, and less chance that it will do the job precisely.

But with the a7 cameras inflicting the bit-depth tax when you use continuous auto-bracketing, you're going to lose about a stop of DR just for invoking that mode. So, if you select as a bracketing sequence 0, +0.3, -0.3 stops, you're actually going to be worse off than if you'd not used auto-bracketing in the first place.

To my way of thinking, the minimum continuous auto-bracketing span that makes any sense is 2 stops, and even that is less than a one-stop improvement over no auto-bracketing at all. So use 0, -1, +1 or more when you use continuous auto-bracketing.

Very interesting. Thanks for the info.

 Camley's gear list:Camley's gear list
Canon EOS 6D Sony a7R II
echolsphoto
echolsphoto Senior Member • Posts: 1,047
Re: Continuous bracketing on the alpha 7 cameras

+1

Chrisd999
Chrisd999 Senior Member • Posts: 1,147
Re: Continuous bracketing on the alpha 7 cameras
1

Thanks Jim, this is truly helpful information for those of us who are seeking maximum dynamic range in their photography. More often than not, with my A7R (and now with my A7II) I would shoot off +/-2 stop continuous brackets, although 99% of the time I just pick the best exposure and throw away the other 2 images. Only occasionally do I blend the exposures in PS or HDR them. Based on this, going forward I will work harder (or smarter) to nail the exposure with a single shot, and only where necessary, bracket is single shot mode. Too bad Sony has to cripple the continuous bracketing feature like this.

 Chrisd999's gear list:Chrisd999's gear list
Sony RX1R II Olympus E-M1 II Panasonic Lumix DC-G9 Olympus M.Zuiko Digital ED 12mm 1:2 Samyang 7.5mm F3.5 UMC Fisheye MFT +10 more
JimKasson
OP JimKasson Forum Pro • Posts: 21,718
Re: Continuous bracketing on the alpha 7 cameras
1

Chrisd999 wrote:

Too bad Sony has to cripple the continuous bracketing feature like this.

I think it was not out of malice, but a consequence of the way they implemented continuous drive mode. They wanted maximum speed, so they dropped a bit. Then, when the continuous bracket mode was implemented, they had the choice of adding a new continuous drive mode or re-purposing the one that was already there. The former would have cost time and energy, and the latter was easy. Easy won. Somehow the idea that customers using continuous bracketing would want max DR got lost in the shuffle.

My Nikon cameras don't do this. Not only that, they offer a choice of frame rates in continuous mode. If Sony offered a choice of rates, they could have 13-bit slow ones and 12-bit fast ones.

But my biggest complaint about this is that Sony hasn't documented the way their cameras work in this respect. Customers have to figure it out for themselves. They did the same with raw compression, and I don't think the secrecy approach has worked well for them there, either.

Jim

-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
exapixel Contributing Member • Posts: 543
Re: Continuous bracketing on the alpha 7 cameras
1

Please help me understand this reduction in bit depth a little better.

The Sony "raw" file conveys 11-bit pixel values. These codes are expanded into 14-bit values by a raw converter. So while there are 16,384 possible values for each pixel, only 1793 are ever used (that's 2048, less the 255 redundant values for zero). These 1793 possible tones are distributed logarithmically over the 16K space, and are relatively dense in the darker tones (every other value) and very thin in the brighter tones (one out of 32).

When the camera degrades to a smaller bit depth in continuous advance mode, how is it manifested in the "raw" file? Does the mapping to the 14-bit space in the raw converter look thinner (i.e., 1/4 of dark tones are available, 1/64 of bright)? Does the tonal range drop down to 1024-255=769 possibilities?

I often ETTR, almost always with bracketing (since the in-camera histograms are not accurate) so that I can keep the one exposure that got closest to blowing out.

-- hide signature --

"I like your pictures. You must have a nice camera!" -- "Thanks. I like your poetry. You must have a nice pen."

 exapixel's gear list:exapixel's gear list
Nikon D800E Nikon AF-S Nikkor 85mm f/1.4G Sigma 35mm F1.4 DG HSM Art Sony RX100 II Sony Alpha NEX-7 +6 more
JimKasson
OP JimKasson Forum Pro • Posts: 21,718
Re: Continuous bracketing on the alpha 7 cameras
1

exapixel wrote:

Please help me understand this reduction in bit depth a little better.

The Sony "raw" file conveys 11-bit pixel values. These codes are expanded into 14-bit values by a raw converter. So while there are 16,384 possible values for each pixel, only 1793 are ever used (that's 2048, less the 255 redundant values for zero). These 1793 possible tones are distributed logarithmically over the 16K space, and are relatively dense in the darker tones (every other value) and very thin in the brighter tones (one out of 32).

When the camera degrades to a smaller bit depth in continuous advance mode, how is it manifested in the "raw" file? Does the mapping to the 14-bit space in the raw converter look thinner (i.e., 1/4 of dark tones are available, 1/64 of bright)?

I often ETTR, almost always with bracketing.

Near as I can tell from photon transfer curves and histogram analysis, the linear a7x raw files are never 14 bits. In continuous drive mode, they are 12 bits, and in single shot drive mode they are 13 bits. Then a tone curve is applied and the bit to take the 13 bit representation to 11 bits. If the linear representation is 12 bits, the tone curve is applied as if the representation were 13 bits with the last bit always being zero.

Does that make sense to you?

Jim

-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
exapixel Contributing Member • Posts: 543
Re: Continuous bracketing on the alpha 7 cameras
1

JimKasson wrote:

exapixel wrote:

Please help me understand this reduction in bit depth a little better.

The Sony "raw" file conveys 11-bit pixel values. These codes are expanded into 14-bit values by a raw converter. So while there are 16,384 possible values for each pixel, only 1793 are ever used (that's 2048, less the 255 redundant values for zero). These 1793 possible tones are distributed logarithmically over the 16K space, and are relatively dense in the darker tones (every other value) and very thin in the brighter tones (one out of 32).

When the camera degrades to a smaller bit depth in continuous advance mode, how is it manifested in the "raw" file? Does the mapping to the 14-bit space in the raw converter look thinner (i.e., 1/4 of dark tones are available, 1/64 of bright)?

I often ETTR, almost always with bracketing.

Near as I can tell from photon transfer curves and histogram analysis, the linear a7x raw files are never 14 bits. In continuous drive mode, they are 12 bits, and in single shot drive mode they are 13 bits. Then a tone curve is applied and the bit to take the 13 bit representation to 11 bits. If the linear representation is 12 bits, the tone curve is applied as if the representation were 13 bits with the last bit always being zero.

Does that make sense to you?

Completely. Thanks.

-- hide signature --

"I like your pictures. You must have a nice camera!" -- "Thanks. I like your poetry. You must have a nice pen."

 exapixel's gear list:exapixel's gear list
Nikon D800E Nikon AF-S Nikkor 85mm f/1.4G Sigma 35mm F1.4 DG HSM Art Sony RX100 II Sony Alpha NEX-7 +6 more
MingTyh Maa Regular Member • Posts: 162
Re: Continuous bracketing on the alpha 7 cameras

Is this true for the A6000 as well?

 MingTyh Maa's gear list:MingTyh Maa's gear list
Sony RX10 III Sony RX100 V Sony a7R II Sony a7R III Sony Vario-Tessar T* FE 16-35mm F4 ZA OSS +2 more
JimKasson
OP JimKasson Forum Pro • Posts: 21,718
Re: Continuous bracketing on the alpha 7 cameras

MingTyh Maa wrote:

Is this true for the A6000 as well?

I don't know, and I don't have an A6000 to test.

Sorry,

Jim

-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
Jerry Fusselman
Jerry Fusselman Contributing Member • Posts: 849
Re: Continuous bracketing on the alpha 7 cameras
1

JimKasson wrote:

<snip>

To my way of thinking, the minimum continuous auto-bracketing span that makes any sense is 2 stops, and even that is less than a one-stop improvement over no auto-bracketing at all. So use 0, -1, +1 or more when you use continuous auto-bracketing.

-- hide signature --

Jim, thanks so much for your vast research on this and many other areas relating to the A7 family.  However, this last paragraph seems to me to contain a fallacy.  You seem to be equating the value of the least-significant bit to the most significant.

For example, imagine a scene with noncrucial shadow detail---at least noncrucial on a wide-DR camera like the A7r at ISO 100.  It is quite likely that an EV0 raw file of this scene may have totally unimportant least-signficant bits.  One simple way to prove it would be to reduce exposure by one stop using the L channel of the Lab color space, and then add the stop back in Lab and see if a majority prefers the original.

In general, it is a mistake to use ACR's methods for changing exposure, for they have the pointless flaw of using a master-curve approach that simultaneously messes up colors and increases noise.  I realize that this goes against conventional wisdom, but see, for example, the last two figures of chapter 14 of Dan Margulis's MPCW (2013).

(By the way, this common flaw of relying on ACR for exposure changes exists in all tests for ISOlessness that I have yet found, though I'll keep looking.  Maybe I should make the time to do my own tests.)

Back to my example of an image that has no crucial shadow detail in which the least-significant bit (when shot at EV0) is irrelevant to the image itself.  In this case, exposure compensation of +1 or -1 could dramatically improve the image through better placement of highlight detail, despite the loss of a bit.

Indeed, I would expect this to be the case for most landscape situations, for example.  I would expect most landscape shooters to get better results with continuous auto-bracketing of 0,-1,+1 than with a single shot with that extra bit.  I am even ignoring the possibility of HDR.  I believe that I have results to back that up.

Back again to the hypothetical:  Due to a possible better placement of highlight detail, it is even quite possible that -1/3 or +1/3 with the least-significant bit missing could be considerably better than the +0 version with that bit present.

In summary, I think you are understating the expected value of continuous autobracketing on an A7/r/s/II camera.  (Yet, I admit that you have a point about that last bit having _some_ value, and I often use single autobracketing when I have time and I am using a tripod.)

Jerry

 Jerry Fusselman's gear list:Jerry Fusselman's gear list
Sony a7R II Sony a9 Canon TS-E 17mm f/4L Canon TS-E 90mm f/2.8 Canon MP-E 65mm f/2.5 1-5x Macro +21 more
JimKasson
OP JimKasson Forum Pro • Posts: 21,718
Re: Continuous bracketing on the alpha 7 cameras

Jerry Fusselman wrote:

JimKasson wrote:

<snip>

To my way of thinking, the minimum continuous auto-bracketing span that makes any sense is 2 stops, and even that is less than a one-stop improvement over no auto-bracketing at all. So use 0, -1, +1 or more when you use continuous auto-bracketing.

Jim, thanks so much for your vast research on this and many other areas relating to the A7 family. However, this last paragraph seems to me to contain a fallacy. You seem to be equating the value of the least-significant bit to the most significant.

I don't think so. It may be a matter of normalization. If the ADC range is simple unsigned integer representation, then the value of the LSB is always one, and full scale is 2^nBits-1, so adding a bit of signal doubles the engineering dynamic range. If you prefer to normalize to floating point numbers between zero and one, which is the way I usually work, then adding a bit means that the LSB is one-half of what it was before, and you've again doubled the engineering dynamic range.

For example, imagine a scene with noncrucial shadow detail---at least noncrucial on a wide-DR camera like the A7r at ISO 100. It is quite likely that an EV0 raw file of this scene may have totally unimportant least-signficant bits. One simple way to prove it would be to reduce exposure by one stop using the L channel of the Lab color space, and then add the stop back in Lab and see if a majority prefers the original.

If the last few LSBs are unimportant, then, in my mind, it's not a candidate for exposure bracketing. I see you disagree below.

In general, it is a mistake to use ACR's methods for changing exposure, for they have the pointless flaw of using a master-curve approach that simultaneously messes up colors and increases noise.

That is not my opinion.

I've done a study of how Lr's PV2012 exposure controls work, and in general I find them preferable to the way the older PV's worked, and how Exposure works in Ps.

http://blog.kasson.com/?p=2960

http://blog.kasson.com/?p=2994

Back to my example of an image that has no crucial shadow detail in which the least-significant bit (when shot at EV0) is irrelevant to the image itself. In this case, exposure compensation of +1 or -1 could dramatically improve the image through better placement of highlight detail, despite the loss of a bit.

Now we're discussing whether pushing in post in Lr really works as intended. For me, for the most part, if you don't get too carried away, it does. I offer the following examples with Lr pushes for several different cameras:

a7II: http://blog.kasson.com/?p=8365

D810: http://blog.kasson.com/?p=6574

a7S high ISO: http://blog.kasson.com/?p=6267

a7S low ISO: http://blog.kasson.com/?p=6241

a7: http://blog.kasson.com/?p=4975

a7R: http://blog.kasson.com/?p=3826

Those all work pretty well, but there are problems with Lr pushes on M240 files at ISOs below 100.

M240: http://blog.kasson.com/?p=3161

M240 shadows: http://blog.kasson.com/?p=3172

I blame the camera's improper black point subtraction for this, not Lr. By the way, there is a Lr plugin that fixes the problem:

http://blog.kasson.com/?p=6717

I admit that there are "twists" in some Lr profiles, but I've not found them to be prolematical for the most part.

Jim

-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon D5 Fujifilm GFX 50S Sony a9 Sony a7R III Sony a7 III +5 more
Parentesi New Member • Posts: 18
Re: Continuous bracketing on the alpha 7 cameras

Will the new A7R not have this "issue"?

I mean we are talking about three or 5 continuous shots. What time-benefit do I have? And why does Sony not tell us this?

I am quite confuses.

I did test this, and let my panoramic-head shot in continuous Bracketing, mens the head is releasing the camera once for all the shots and camera set to 3BR and in single Bracketing mode, the head shooting three signal per posting. So in the real dark areas and in the underexposed images, I can really see a big difference when pushing this up.

The single-shot remain some texture, the continuous showing bigger blemish areas of just one color "green"

 Parentesi's gear list:Parentesi's gear list
Sony a7R
Liske New Member • Posts: 1
Re: Continuous bracketing on the alpha 7 cameras

What would be the easiest way to fore off a bracket sequence at 13 bit? bulb? iPhone?

JimKasson wrote:

Automatic bracketing is implemented on the Sony alpha 7 cameras as a shutter, or drive, mode,. There are two modes, continuous and single shot. If you use continuous mode, the raw (pre-compression) bit depth is changed from 13 bits to 12 bits, just like when the camera operates in continuous shutter mode without auto-bracketing.

If you're doing handheld HDR, you want continuous auto-bracketing. If you use single shot auto-bracketing, there will be greater camera motion between shots, more work for the auto-registration feature of your HDR program to do, and less chance that it will do the job precisely.

But with the a7 cameras inflicting the bit-depth tax when you use continuous auto-bracketing, you're going to lose about a stop of DR just for invoking that mode. So, if you select as a bracketing sequence 0, +0.3, -0.3 stops, you're actually going to be worse off than if you'd not used auto-bracketing in the first place.

To my way of thinking, the minimum continuous auto-bracketing span that makes any sense is 2 stops, and even that is less than a one-stop improvement over no auto-bracketing at all. So use 0, -1, +1 or more when you use continuous auto-bracketing.

Keyboard shortcuts:
FForum MMy threads