Sony A7r sensor not the same

[Ben Rubinstein]

My take, using a 50 year old Pentax 50mm with a scratched rear element @ f8 (2 series gitzo, BH-40 2 second delay on shutter), is that the A7r is sharper and far more detailed than my 5D3 ever was with any of my L lenses. I can't comment on clarity and contrast due to the lens which is yellowed due to a radioactive element. We have a D800e in the studio and shooting repro would be a great way to test this but to be honest, life it rather too short to bother, we've got too much real work to do, maybe another day less busy than this coming week (have to take the DF to the dealer tomorrow for 2nd shutter replacement in 6 months, sigh). I stand by my impression reported at the beginning. The raw file is here, focus on the AC unit. It's a DNG using the same base (LR/ACR) settings I had for my 5D3 and custom sharpening (I had to back the sharpening down from my defaults it was so sharp). It was this file that made me decide not to bother with a 'lux or a Sonnar. I love what that lens does wide open and stopped down it's bleeding sharp. Only thing is the Adobe jaggies are very apparent in that shot. Was the first time I'd seen the problem.

EDIT. Just tried it with C1. Doesn't sharpen up anywhere near as nicely as in ACR. Tried it with 210,0.6,0 but the clarity sucks at default unlike in Adobe and the lack of a 'detail' slider is a real problem when trying to sharpen up brick walls. Jaggies are still there but less obvious. Try this in Adobe people.

 

[dchew]

Jack,
There are two other possibilities:
1. It's been probably two years since I've tested stuff. I use LR 80% of the time, and like many people after a while I get to know the sharpening required, so I use presets that are applied during import. Frankly it's been a while since I've looked at unsharpened images. Maybe I'm just expecting too much.

2. It is the compression. Your comment about aliasing is interesting, because in the one image below of my house (#465), I see aliasing around the center window frames. That tells me it's focused pretty good. So how could it be a bit fuzzy but still have aliasing?

#430 is the 90ts at f/8. Sorry I guess I uploaded the wrong one!
#455 is the 35 f/2 at f/4
http://www.hightail.com/download/elNLL0dObThVbSt4djhUQw

#465 is the 35f/2 at f/4; house outside maybe a bit more relevant to the "high frequency" discussion. Focus on the four bricks above the center windows
http://www.hightail.com/download/elNLL0dEayt0QTFFQmNUQw

Get rid of shutter vibration.

http://www.getdpi.com/forum/558875-post139.html post #139

KH, thank you. I've seen your great work and posts. But honestly I'm seeing the same thing in your images.

I want to be clear: this is not something I'm disappointed with! It is completely eliminated with a bit of standard sharpening, CA removal, etc. I'm just curious why I am noticing the same thing as Jack. Although I don't have a D800e to compare it to.

Dave

 

[k-hawinkler]

Jack,
There are two other possibilities:
1. It's been probably two years since I've tested stuff. I use LR 80% of the time, and like many people after a while I get to know the sharpening required, so I use presets that are applied during import. Frankly it's been a while since I've looked at unsharpened images. Maybe I'm just expecting too much.

2. It is the compression. Your comment about aliasing is interesting, because in the one image below of my house (#465), I see aliasing around the center window frames. That tells me it's focused pretty good. So how could it be a bit fuzzy but still have aliasing?

#430 is the 90ts at f/8. Sorry I guess I uploaded the wrong one!
#455 is the 35 f/2 at f/4
http://www.hightail.com/download/elNLL0dObThVbSt4djhUQw

#465 is the 35f/2 at f/4; house outside maybe a bit more relevant to the "high frequency" discussion. Focus on the four bricks above the center windows
http://www.hightail.com/download/elNLL0dEayt0QTFFQmNUQw




KH, thank you. I've seen your great work and posts. But honestly I'm seeing the same thing in your images.

I want to be clear: this is not something I'm disappointed with! It is completely eliminated with a bit of standard sharpening, CA removal, etc. I'm just curious why I am noticing the same thing as Jack. Although I don't have a D800e to compare it to.

Dave

Could you please be more specific?
I have lots of images also with shutter shake.
How about this one?
Leica User Forum - Einzelnen Beitrag anzeigen - A7[R] Images with Leica lenses (open thread)
Taken with A7R. I think it's okay.

Or this one.
http://www.getdpi.com/forum/559634-post149.html
Taken with NEX-7.

These are full resolution images, files.
So, a computer is probably required to inspect them.

 

[Bob]

Curiouser and curiouser...
I was poking around the innards of an A7r "raw" file and came across something interesting that may explain a few things including why it seems to need so much more sharpening than one might otherwise expect.

For reference, almost all raw files are simply tiffs really, and can be examined by common tools.
Within each file, there are tags that indicate to the raw processor, some information needed so it can do its job properly.
Below are some tags from a D800 file for comparison:
SubFileType (1 Long): ReducedImage
ImageWidth (1 Long): 160
ImageLength (1 Long): 120
BitsPerSample (3 Short): 8, 8, 8
Compression (1 Short): Uncompressed
Photometric (1 Short): RGB
Make (18 ASCII): NIKON CORPORATION
Model (12 ASCII): NIKON D800E
StripOffsets (1 Long): 129748
Orientation (1 Short): TopLeft
SamplesPerPixel (1 Short): 3
RowsPerStrip (1 Long): 120
StripByteCounts (1 Long): 57600
XResolution (1 Rational): 300
YResolution (1 Rational): 300
PlanarConfig (1 Short): Contig
ResolutionUnit (1 Short): Inch
Software (10 ASCII): Ver.1.02
DateTime (20 ASCII): 2013:09:25 22:20:17
Artist (37 ASCII): Bob Freund
SubIFD (3 Long): 187348, 187468, 187696
ReferenceBlackWhite (6 Rational):
700 (1024 Byte): 60, 63, 120, 112, 97, 99, 107, 101, 116, 32,...
Copyright (55 ASCII): Copyright 2012 Bob Freund ...
34665 (1 Long): 1640
34853 (1 Long): 129728
36867 (20 ASCII): 2013:09:25 22:20:17
37398 (4 Byte): 1, 0, 0, 0

note that image tile itself is uncompressed. Nikon achieves its lossless compression by profiling the image color space and mapping the set of colors discovered into a smaller number of encoded bits (if possible)

Now for the tags from a sample A7r raw file:
SubFileType (1 Long): ReducedImage
Compression (1 Short): JPEG Original TIFF 6.0 spec
ImageDescription (32 ASCII):
Make (5 ASCII): SONY
Model (8 ASCII): ILCE-7R
Orientation (1 Short): TopLeft
XResolution (1 Rational): 350
YResolution (1 Rational): 350
ResolutionUnit (1 Short): Inch
Software (14 ASCII): ILCE-7R v1.01
DateTime (20 ASCII): 2014:01:01 14:29:44
SubIFD (1 Long): 143834
JpegIFOffset (1 Long): 144546
JpegIFByteCount (1 Long): 969783
YCbCrPositioning (1 Short): 2
34665 (1 Long): 432
50341 (106 Undefined):
50740 (4 Byte): 228, 194, 0, 0

Note that the image tile is compressed using the jpeg compression algorithm according to the TIFF 6.0 spec (1992)

That spec provides for both lossless and lossy compression possibilities (Section 22) and I have not peeled the onion an additional layer to see what transform they are using, but it does appear that one of the jpeg compression techniques is used in the Sony raw files. According to that spec, all but one algorithm is lossy, and the only non-lossy algorithm is one that uses Huffman encoding.
I will dig a bit further to see what I can find out. The bits will tell the truth.
-bob

 

[k-hawinkler]

Many thanks Bob for leading the way.
Much appreciated. Very interesting indeed!

 

[sergio lovisolo]

Here is an example of aliasing with the A7r and a summicron m last type (non apo) at 5,6.
The file has been sharpened in ACR8 at 60/0,6/25/0


_DSC0597 by sergio lovisolo, on Flickr

What I see here:

The file is over sharpened, and this is the reason for which i never apply sharpening to the A7r files, unless this is required by limits of the lens used.

The camera shows the correct resolution for 36 Mp, with total pixel sharpness, while the lens has a lot more high frequency energy to supply to the sensor.

It is impossible to obtain better results resolution wise by 36 Mp sensors.

I am completely satisfied by the quality of the files, and don't expect anything better.

Sergio

 

[fotografz]

I think Sony is experimenting with different concepts with their Bionz X chip which they in part call "Digital Reproduction Technology".

"Processor. Output from the Sony A7R's image sensor is handled by a brand-new BIONZ X-branded image processor. Compared to the previous generation, BIONZ X has improved performance. Sony also says that it has improved area-specific noise reduction, which varies noise reduction strength across the image in an attempt to yield a clean result without disturbing image detail.

BIONZ X also brings two new features. Diffraction-reducing technology combats the effects of diffraction limiting, improving detail at smaller apertures. Interestingly, the effect applied is both lens and aperture specific, and said to work even with Alpha-mount lenses shot through an adapter. It also has what Sony bills as "Detail reproduction technology", which tries to draw out finer details without creating halos in the process." _-DP Review (Same thing has been said on other reviews). DP Review hasn't varivied any of this, so ...

Blah, blah, blah ... marketing speak ... BUT:

It seems to imply that what they are doing is more specific to their data bus lenses (FE and A Mount). Which makes sense because the Sony LAEA-3&4 adapters are brand new just for this camera and have the data bus connections ... what we do not know is if they differ from 3rd party ones that may include data bus connections to use lenses on the NEX cameras. I can say that the A mount ZAs do very well on this camera with the LAEA-4 adapter. Very well.

The notion of less dependance on edge sharpening verse the over-all "impression" of image sharpness reminds me of "Micro Contrast", and resulting debates/discussions. Whatever they are doing to provide that appearance of acuity without the creation of halos may be why there is less tendency to moiré ... and perhaps why some folks feel compelled to sharpen when viewing on screen at 100%+.

My impression so far has been that the over-all image looks and feels more organic and doesn't require any Herculean effort in post ... quite the contrary, these are very easy to PP. This impression is based on prints as well as on screen ... however, I can see that some may wish nothing done to an image so they can do as they wish. I'm not one of those people and never have been (thus the switch from the D3X to the A900). The base image from the A7R still leaves plenty of creative latitude and malleability to play with IMHO.

The other achievement that I like is the stabilization of color and skin tones as one moves up the ISO scale (something my pal Irakly marveled at also). Perhaps not that important to some, of paramount importance to me.

Here is an example: (pardon the repetitious "Sleeping Grandma" subject matter, but Ive been stuck here with family )

Basically, it printed beautifully on 13" X 19" paper, with a nice sense of depth, faithful color, and organic sense of detail at the point of focus ... and while the extreme detail crops on screen could be viewed as needing sharpening, the print says otherwise.

Personally, I think we are dealing with some new ideas here, and need to be patient while learning whether we like it or not. So far, I do. Others may not.

A7R, ISO 400, 1/80 shutter, FE55/1.8 @ f/4 Processed in LR4:

- Marc

 

[Bob]

One more peel of the onion...
The jpeg encoding is huffman so it should not be lossy although it can be if precessed in-camera. It ought to be good quality unless the math is screwed up.
The Bayer array is a conventional RGGBRGGB...
The non-debayered image is a yucky green as is usual,
There is no embedded icc camera profile.

So the test is to use the damned thing and if you like it, then great.
Every camera is subject to close and perhaps over-examination.
Personally, I have a lot of cash in what I have, and unless there is a striking improvement I do not plan on changing. I do not see a striking improvement.
On the positive note, the files can be easily converted to tiffs using dcraw which is available as a c code distribution for those of you who like to play in the command-line space.
-bob

my usual raw file play tools are:
dcraw Decoding raw digital photos in Linux
exiftool ExifTool by Phil Harvey
and of course your favorite xml editor

 

[fotografz]

One more peel of the onion...
The jpeg encoding is huffman so it should not be lossy although it can be if precessed in-camera. It ought to be good quality unless the math is screwed up.
The Bayer array is a conventional RGGBRGGB...
The non-debayered image is a yucky green as is usual,
There is no embedded icc camera profile.

So the test is to use the damned thing and if you like it, then great.
Every camera is subject to close and perhaps over-examination.
Personally, I have a lot of cash in what I have, and unless there is a striking improvement I do not plan on changing. I do not see a striking improvement.
On the positive note, the files can be easily converted to tiffs using dcraw which is available as a c code distribution for those of you who like to play in the command-line space.
-bob

my usual raw file play tools are:
dcraw Decoding raw digital photos in Linux
exiftool ExifTool by Phil Harvey
and of course your favorite xml editor

Bingo!

I have a lot of cash wrapped up in my tools also, and I could not care less if this camera lags, equals or out does my Leica S2. What it does do for sure is allow me to travel with less to carry, shoot casually with AF and less to carry ... do some paying work with less to carry, do some interesting things in the studio with improved mag EVF/focus peaking, and so on. Those attributes fall with-in "striking improvements" for what I do these days.

Even if IQ only equaled my Sony A99, it'd still be the choice for the above reasons.

Frankly, IF I didn't already have an array of A mount lenses, I'm not sure I would've paid as much attention to it no matter how revolutionary it may or may not be ... if ALL M lenses had worked well, then it would've been a different story ... however, since I will not pay $7K for color M work (my main M focus is the M Mono for rangefinder work), what M lenses do work on the little Sony will have to do for now ... supplemented by the FEs, including hopefully the 24-70/4 zoom. Camera + 3 lenses, all at a cost less than the M240 body ... with a wad of change left over for a "fun with" adventure.

Fun times for sure.

- Marc

 

[dchew]

Could you please be more specific?
I have lots of images also with shutter shake.
How about this one?
Leica User Forum - Einzelnen Beitrag anzeigen - A7[R] Images with Leica lenses (open thread)
Taken with A7R. I think it's okay.

Or this one.
http://www.getdpi.com/forum/559634-post149.html
Taken with NEX-7.

These are full resolution images, files.
So, a computer is probably required to inspect them.

I downloaded the raw file from the first link. Yes, I see the same thing there. It is so darn hard to tell in most of these images which is why Bob, Marc and others are exactly right: Too much analysis...

If I zoom into your raw file, the trees on the top ridgeline have haze and climate impacts that lose detail. But somewhere down closer to the bottom, where it looks like the plane of focus was, has what I'm talking about. I can't tell from the other getDPI link because the jpeg makes it look a bit plastic up close and personal.

I went back and looked at another test I did, comparing my IQ180/sk150 with the a7r/Canon 70-200 f/4 IS set at 93mm to get close to the sk150's coverage. Unfair? Well maybe not so much. That combination looks very comparable. Of course the IQ180 file is zoomed in twice as far at 100% because of the additional pixels, but the amount of capture sharpening I add to get it sharp to my taste is actually less on the a7r. My 70-200 is wonderful; almost as good as the 135 f/2.

So In my case what I'm seeing (or imagining) is lens dependent. I don't think it is from the shutter. I think either I've gone in the weeds, or I am seeing lens variation at levels I couldn't before in a 135 format. Doesn't explain Jack's impression though.

Dave

 

[GrahamWelland]

One more peel of the onion...

....

So the test is to use the damned thing and if you like it, then great.
Every camera is subject to close and perhaps over-examination.

-bob

Sage words from bob.

I'm thinking that this should be the banner on all of these forums! :thumbs:

 

[claymore]

One more peel of the onion...
The jpeg encoding is huffman so it should not be lossy although it can be if precessed in-camera. It ought to be good quality unless the math is screwed up.
The Bayer array is a conventional RGGBRGGB...
The non-debayered image is a yucky green as is usual,
There is no embedded icc camera profile.

So the test is to use the damned thing and if you like it, then great.
Every camera is subject to close and perhaps over-examination.
Personally, I have a lot of cash in what I have, and unless there is a striking improvement I do not plan on changing. I do not see a striking improvement.
On the positive note, the files can be easily converted to tiffs using dcraw which is available as a c code distribution for those of you who like to play in the command-line space.
-bob

my usual raw file play tools are:
dcraw Decoding raw digital photos in Linux
exiftool ExifTool by Phil Harvey
and of course your favorite xml editor

Bob,
The JPEG data you posted earlier applies to the preview image embedded in the RAW. The a7/R files are in ARW 2.3 format and the dcraw function that decompresses the version 2 format is called sony_arw2_load_raw. If you look at the code in http://www.cybercom.net/~dcoffin/dcraw/dcraw.c it is not doing JPEG decompression. It is a Sony-specific compression scheme, the basic summary of which is that each pixel value is mapped down from 14 bits to 11, then compressed down to 8 bits per pixel. It is a fixed-size compression scheme so the total size of the compressed raw data will always be 8 bits (1 byte) per pixel regardless of image content.

Here is the main body of the decompression function with my comments added to explain what it is doing:

  for (row=0; row < height; row++) {
    fread (data, 1, raw_width, ifp);
    for (dp=data, col=0; col < raw_width-30; dp+=16) {   // decompress 16 pixels at a time; dp is the start of the current 16 pixels; dp+=16 advances by 16 bytes, which is one byte or 8 bits per pixel
      max = 0x7ff & (val = sget4(dp));                   // read the 11 bit maximum value for these 16 pixels
      min = 0x7ff & val >> 11;                           // read the 11 bit minimum value
      imax = 0x0f & val >> 22;                           // read the position of the maximum-valued pixel (4 bits is enough for the 16 possible positions)
      imin = 0x0f & val >> 26;                           // read the position of the minimum-valued pixel
      for (sh=0; sh < 4 && 0x80 << sh <= max-min; sh++); // calculate additional shift required for the 7 bit differences between min and max in case the difference between max and min is larger than 7 bits
      for (bit=30, i=0; i < 16; i++)                     // we've used 30 bits above (11x2+4x2) so start the bit counter at 30; now we will loop 16 times; only 14 pixels will have values here, as the min/max values are already known from above
	if      (i == imax) pix[i] = max;                // if this is the maximum pixel's position, just assign it the value from above and don't advance the bit counter
	else if (i == imin) pix[i] = min;                // otherwise if this is the minimum pixel's position just assign it the value from above and don't advance the bit counter
	else {                                           // otherwise for the other 14 pixels out of the 16....
	  pix[i] = ((sget2(dp+(bit >> 3)) >> (bit & 7) & 0x7f) << sh) + min;  // read the 7 bit value for the difference between this pixel and the 11 bit minimum, apply the additional shift if any to that difference, then add it to the 11 bit minimum to get this pixels 11 bit value
	  if (pix[i] > 0x7ff) pix[i] = 0x7ff;            // if the shift pushed it past 11 bits, just set it the be the 11 bit maximum value
	  bit += 7;                                      // advance the bit counter by 7 bits
	}
      for (i=0; i < 16; i++, col+=2)
	RAW(row,col) = curve[pix[i] << 1] >> 2;          // apply the 11 to 14 bit curve to each the 16 pixels read above
      col -= col & 1 ? 1:31;
    }
  }

 

[k-hawinkler]

I downloaded the raw file from the first link. Yes, I see the same thing there. It is so darn hard to tell in most of these images which is why Bob, Marc and others are exactly right: Too much analysis...

If I zoom into your raw file, the trees on the top ridgeline have haze and climate impacts that lose detail. But somewhere down closer to the bottom, where it looks like the plane of focus was, has what I'm talking about. I can't tell from the other getDPI link because the jpeg makes it look a bit plastic up close and personal.

I went back and looked at another test I did, comparing my IQ180/sk150 with the a7r/Canon 70-200 f/4 IS set at 93mm to get close to the sk150's coverage. Unfair? Well maybe not so much. That combination looks very comparable. Of course the IQ180 file is zoomed in twice as far at 100% because of the additional pixels, but the amount of capture sharpening I add to get it sharp to my taste is actually less on the a7r. My 70-200 is wonderful; almost as good as the 135 f/2.

So In my case what I'm seeing (or imagining) is lens dependent. I don't think it is from the shutter. I think either I've gone in the weeds, or I am seeing lens variation at levels I couldn't before in a 135 format. Doesn't explain Jack's impression though.

Dave

Thanks Dave. I am sorry but I have no idea what you are talking about.
But I really would like to understand. TIA.

 

[BSEH]

Im not the teq guy here, and really having a hard time understand how this wil affect my pictures.. so I try making a head to head - D800 sensor vs the A7r. Same lens, Zeiss 28 mm f2.8, at 5,6. Sensor exposed with flash, so no shutter/or fat fingers to mess thing up. Flash duration 1/11.000 in t0.5 - can you tell who is who ?
150% crop..

 

[Jack]

Neither one looks particularly sharp on my monitor, so it's tough to say which cam. Moreover, the top image is a good 1/3rd stop under-exposed from the bottom image, which essentially negates any worthwhile comparisons of contrast. What I will say is either image seems to be lens limited... That said, the 2nd image (lower frame) appears to be slightly higher resolution, but again that could be a visual effect of the extra 1/3 stop of exposure.

 

[Jack]

Jack,
There are two other possibilities:
1. It's been probably two years since I've tested stuff. I use LR 80% of the time, and like many people after a while I get to know the sharpening required, so I use presets that are applied during import. Frankly it's been a while since I've looked at unsharpened images. Maybe I'm just expecting too much.

2. It is the compression. Your comment about aliasing is interesting, because in the one image below of my house (#465), I see aliasing around the center window frames. That tells me it's focused pretty good. So how could it be a bit fuzzy but still have aliasing?

#430 is the 90ts at f/8. Sorry I guess I uploaded the wrong one!
#455 is the 35 f/2 at f/4
http://www.hightail.com/download/elNLL0dObThVbSt4djhUQw

#465 is the 35f/2 at f/4; house outside maybe a bit more relevant to the "high frequency" discussion. Focus on the four bricks above the center windows
http://www.hightail.com/download/elNLL0dEayt0QTFFQmNUQw

I'm leaning toward compression. At least the image of your window looks sharp centrally, and does show a hint of moire in the window moldings. But I gave your image the same rather heavy-handed adjustments I mentioned earlier and it holds them well, suggesting something is going on internally.

 

[BSEH]

Neither one looks particularly sharp on my monitor, so it's tough to say which cam. Moreover, the top image is a good 1/3rd stop under-exposed from the bottom image, which essentially negates any worthwhile comparisons of contrast. What I will say is either image seems to be lens limited... That said, the 2nd image (lower frame) appears to be slightly higher resolution, but again that could be a visual effect of the extra 1/3 stop of exposure.

Dam.. Guess I moved the flash - on more try.. Forgot to tell I was kick out of Physic Class - Humaniore is my speciality

 

[dchew]

I should apologize to everyone. I went back and looked at more old 5DII images with sharpening disabled, and they are not any better than what I am seeing with the a7r at 100%.

I suspect what I am seeing is either simply the difference between very expensive wide angle Rodenstock glass, or what Jack suggests associated with compression.

Regardless, I am thrilled with this camera and its image quality.

Dave

 

[Don Libby]

One more peel of the onion...

So the test is to use the damned thing and if you like it, then great.
Every camera is subject to close and perhaps over-examination.

-bob

I'm in total agreement with Bobs statement. It seems there is this same type of discussion every time there's a new camera. Best advise is to try it if it works for you then great if not oh well. I don't subscribe to the idea of dissecting each and everything to its core trying to see how it works. Right now until something better comes along it works well for Sandy and me. I did ask Sandy her thoughts and the only negative was the shutter sound then she also said she's gotten used to it. Bottom line it works for us.

I'm cranky, tired and cold but thought I'd get this off my chest.....

 

[Deleted member 7792]

Regardless, I am thrilled with this camera and its image quality.

I'm in total agreement with Bobs statement. It seems there is this same type of discussion every time there's a new camera. Best advise is to try it if it works for you then great if not oh well. I don't subscribe to the idea of dissecting each and everything to its core trying to see how it works. Right now until something better comes along it works well for Sandy and me. I did ask Sandy her thoughts and the only negative was the shutter sound then she also said she's gotten used to it. Bottom line it works for us.

I'm cranky, tired and cold but thought I'd get this off my chest.....

Dave & Don, I agree with you both. This little camera has exceeded my expectations. The performance-to-price ratio is remarkable, and my aging back will be happy. In 1998 I was using a Sony Mavica, and after many years of Nikon, Hasselblad, Phase One, Alpa, Cambo and Leica, I'm back to a Sony. Fun times we live in.

Joe