Sensor+ vs. legacy fat pixel backs

[PWS]

hey guys,

having heard/read lots of positive comments about the 'look' of fat pixel backs,
i wonder if Sensor+ mode produces similar results.

I know that in Sensor+ mode 4 pixels are put together, which actually
makes them larger. However, do you also get the same look?

[Ken R]

hey guys,

having heard/read lots of positive comments about the 'look' of fat pixel backs,
i wonder if Sensor+ mode produces similar results.

I know that in Sensor+ mode 4 pixels are put together, which actually
makes them larger. However, do you also get the same look?

Good question. I don't know for sure. But on my IQ160 I have used the sensor+ mode a bunch of times for both product and people shoots. The images have a really nice quality to them, they look a bit smoother and color is just superb even when going into the iso 400 range. Even though the file size on the IQ160 is 15mp in that mode the images look MUCH better than files from DSLRs with similar resolution or even more. Sensor + is one of those things that I though I would never use or liked but have ended up loving it and using it a lot more than I though I would and not just for the cleaner high iso performance. For me its like having two backs in one.

[Steve Hendrix]

Good question. I don't know for sure. But on my IQ160 I have used the sensor+ mode a bunch of times for both product and people shoots. The images have a really nice quality to them, they look a bit smoother and color is just superb even when going into the iso 400 range. Even though the file size on the IQ160 is 15mp in that mode the images look MUCH better than files from DSLRs with similar resolution or even more. Sensor + is one of those things that I though I would never use or liked but have ended up loving it and using it a lot more than I though I would and not just for the cleaner high iso performance. For me its like having two backs in one.

What Ken says echoes my feelings as well about the files. But also that users tend to take advantage of Sensor Plus more than they thought they would before they purchased. There's definitely something to be said for the right tool for the job, but we use DSLR cameras in as many situations as we can, why not do the same with your medium format digital camera system when the opportunity presents itself?


Steve Hendrix
Capture Integration

[Paul2660]

I believe your "fat" sensor backs at base iso will give a better result than say a IQ160, P65+ etc. in sensor plus mode at iso 200 and up.

Sensor plus to me is a "Last" resort, to gain workable shutter speed in the field.  Signal noise is better especially in the shadows, highlights are more prone to blowing out, but the overall loss in resolution from 60MP to 15MP is a huge issue to me.

I wouldn't consider the purchase of a 10K to 15K back for primarily using sensor plus, or with that as I highly listed advantage, mainly due to the 1/4 resolution output.   

I also don't feel from my work with various  16MP 35mm CMOS cameras that the 15MP output from a IQ160 or 260 (both of which I have owned or own) is superior to the 16MP output of a modern CMOS 35mm camera at iso 200, 400 or 1600 for that matter.  The optics I use on the MFD with a tech camera are better but uprezing is uprezing and getting those files back to around 50MP equivalent is pretty hard to do. 

The images from sensor plus are good, but no better than Fuji, Nikon (in DX mode). 

Where the CCD backs are strongest from my work, (outdoor landscape) is base iso, i.e. 50 in this case.  There the output is outstanding, and I do prefer it to a CMOS camera, but pushing them, either in base iso or in sensor plus mode is pretty harsh compromise, (to me).

Paul

[Theodoros]

hey guys,

having heard/read lots of positive comments about the 'look' of fat pixel backs,
i wonder if Sensor+ mode produces similar results.

I know that in Sensor+ mode 4 pixels are put together, which actually
makes them larger. However, do you also get the same look?

Well... My opinion is that there is no "fat pixel magic" onto the Dalsa 22mp sensor... actually I believe that with Dalsa the 33mp sensor is more "magical" than the 22mp one and maybe the most "accurate" sensor that has ever been around... Kodak 22mp though it's a completely different dog... it's not "accurate" but it can provide a look that is better than film in film territory and it can flutter colour and DR in a unique way that no other sensor ever could... No, sensor+ doesn't remind of "fat pixel magic"... it's no where near actually... it is far more accurate and much worst because it will measure better in every respect... (if you know what I mean). 

I tell you what though... most people that sold their Kodak 22mp sensor MFDB they now wish they had it back... Why? you may ask... because they can't have the magic they lost back!!!! ...as simple as that!!!!    I don't think you'll have anyone that have used the Kodak 22mp sensor not supporting what you just read or to tell you "bad" on it... you know a lier when you meet one... 

[eronald]

Well... My opinion is that there is no "fat pixel magic" onto the Dalsa 22mp sensor... actually I believe that with Dalsa the 33mp sensor is more "magical" than the 22mp one and maybe the most "accurate" sensor that has ever been around... Kodak 22mp though it's a completely different dog... it's not "accurate" but it can provide a look that is better than film in film territory and it can flutter colour and DR in a unique way that no other sensor ever could... No, sensor+ doesn't remind of "fat pixel magic"... it's no where near actually... it is far more accurate and much worst because it will measure better in every respect... (if you know what I mean). 

I tell you what though... most people that sold their Kodak 22mp sensor MFDB they now wish they had it back... Why? you may ask... because they can't have the magic they lost back!!!! ...as simple as that!!!!    I don't think you'll have anyone that have used the Kodak 22mp sensor not supporting what you just read or to tell you "bad" on it... you know a lier when you meet one... 

Can you list  fat pixel back models & sensor types?

Edmund

[Theodoros]

Can you list  fat pixel back models & sensor types?

Edmund

What you mean? ..include all the 11mp, 16mp, 20mp & 22mp of the Kodak 9mμ pixel variants?  ...why should I?  What kind of a Q is that? 

[eronald]

What you mean? ..include all the 11mp, 16mp, 20mp & 22mp of the Kodak 9mμ pixel variants?   ...why should I?   What kind of a Q is that?  

You could give the commercial names of the ones you talk about in your post ...
Or say the magic words that summon the appropriate rep and make him reminisce

Edmund

[ErikKaffehr]

Hi,

Sensor+ uses a special method of binning pixels that significantly reduces aliasing, so I would guess that the result will be significantly different from "fat pixels".

It is described here: https://www.captureintegration.com/wp-content/uploads/Phase-One-Sensorplus.pdf

The enclosed image is from the above document. Top represents traditional binning while bottom represents sensor-plus binning.

If you want the large pixel look, you could simply try downscaling a 60 mp image to 15 mp using nearest neighbour. (I actually know nearest neighbour is not the best emulation of large pixels, but it is a much better emulation than bicubic).

All this is oversimplification, of course. Large pixels will give more color aliasing. The suggested method of downsampling a demosaiced image would have much less colour aliasing than a large pixel image. Fill factor also plays a major role.

Best regards
Erik

hey guys,

having heard/read lots of positive comments about the 'look' of fat pixel backs,
i wonder if Sensor+ mode produces similar results.

I know that in Sensor+ mode 4 pixels are put together, which actually
makes them larger. However, do you also get the same look?

[EricWHiss]

I know what Theodoros is speaking about.... my p20 and CF 528 both had the kodak chips with 9um sensor pitch and really they had this look that is just really wonderful, rich with good tonality.  It probably isn't the most correct color, but it still looks great to me.   The Dalsa sensor in the Leaf AFi-ii 12 definitely have a different look to them, maybe all the different sensors have a different look? 

[bcooter]

Sometimes I think when I go over past imagery, I am usually bias.  Maybe it's nostalgia, maybe just because I had more time to work back then, but it's difficult to get a true comparison, except:

We have one on figure fashion project that we have shot essentially the same products, though the product and talent changes, including the lighting and some of the creative brief, the essential concept is usually the same.

We've shot this year to year with the kodak dcs 760, 1ds, 1ds2, 1ds3, aptus 22, p30, p21, 1dx, and Leica S2.

A month or so ago I was on the server and all our selects from this client are in one folder so it's easy to compare and I put all the Canon selects in DPP, the Aptus selects in Adobe camera raw, the phase p30 selects in C-1 and the Leica S2 in lightroom because those are the processors that each respective makers recommend.

If I had to pick the best look, not the most detailed, easy to use, not doing a DR, pixel peeping contest, just comparing out of camera looks with slight tweaks, the overall best looks are:

1.  Original 1ds, 2. Aptus 22, 3+4 tied P20+ and 5 Leica S2.    (The kodak dcs 760 files were very pretty, but at 6mpx lacked the detail needed for this project, but had Kodak continued with this camera to around 11 to 18mpx I'd probably still be using it).

The medium format files are easy to judge because they all were shot with contax Zeiss lenses.

The first 1ds was really pretty, don't know the technical reasons why, just know that the skin tones were even and rich, the product (jeans wear) looked great, the rhythm and flow of the shoot was more natural.

Unlike all the other cameras, including the canons and the S2 the first 1ds files were never over saturated or heavy in the red bias.

The Aptus 22, though different was very pretty, though moire was a real issue.  The P30+, just rocks on then and still does.

The Leica, out of the can with stock lightroom settings is grim, with special calibration applied is very nice, though I wouldn't say it's any better than the older p30+. 

The only issue with C-1 is it's very crunchy sharp out of the can and I guess cause Phase has always dialed in a lot of over sharpening, taken down and working some calibration the p30+ files are the richest of all.

BTW:  Never try S2 files in C-1 as it's so far off in profiles it takes a lot of work to get back to what was viewed on set.

What's funny is the original 1ds is the only camera we didn't tether, we shot to cards and had the client view the project on a half session basis.  I think this is why the flow of the shoot was more natural as there was less "input" as we worked.

IMO

BC

[eronald]

J,

I'm not feeling very smart this morning - does this mean  the only CMOS to make the (your) top was the original 1Ds?

I still have mine. I always hoped  I would get another cam with get the same look and ergonomics, but more pixels. I guess the laws of physics prevent that from happening.

Some more posts like this one and we'll officially be the pensioner's old camera look society.

Edmund

Sometimes I think when I go over past imagery, I am usually bias.  Maybe it's nostalgia, maybe just because I had more time to work back then, but it's difficult to get a true comparison, except:

We have one on figure fashion project that we have shot essentially the same products, though the product and talent changes, including the lighting and some of the creative brief, the essential concept is usually the same.

We've shot this year to year with the kodak dcs 760, 1ds, 1ds2, 1ds3, aptus 22, p30, p21, 1dx, and Leica S2.

A month or so ago I was on the server and all our selects from this client are in one folder so it's easy to compare and I put all the Canon selects in DPP, the Aptus selects in Adobe camera raw, the phase p30 selects in C-1 and the Leica S2 in lightroom because those are the processors that each respective makers recommend.

If I had to pick the best look, not the most detailed, easy to use, not doing a DR, pixel peeping contest, just comparing out of camera looks with slight tweaks, the overall best looks are:

1.  Original 1ds, 2. Aptus 22, 3+4 tied P20+ and 5 Leica S2.    (The kodak dcs 760 files were very pretty, but at 6mpx lacked the detail needed for this project, but had Kodak continued with this camera to around 11 to 18mpx I'd probably still be using it).

The medium format files are easy to judge because they all were shot with contax Zeiss lenses.

The first 1ds was really pretty, don't know the technical reasons why, just know that the skin tones were even and rich, the product (jeans wear) looked great, the rhythm and flow of the shoot was more natural.

Unlike all the other cameras, including the canons and the S2 the first 1ds files were never over saturated or heavy in the red bias.

The Aptus 22, though different was very pretty, though moire was a real issue.  The P30+, just rocks on then and still does.

The Leica, out of the can with stock lightroom settings is grim, with special calibration applied is very nice, though I wouldn't say it's any better than the older p30+.  

The only issue with C-1 is it's very crunchy sharp out of the can and I guess cause Phase has always dialed in a lot of over sharpening, taken down and working some calibration the p30+ files are the richest of all.

BTW:  Never try S2 files in C-1 as it's so far off in profiles it takes a lot of work to get back to what was viewed on set.

What's funny is the original 1ds is the only camera we didn't tether, we shot to cards and had the client view the project on a half session basis.  I think this is why the flow of the shoot was more natural as there was less "input" as we worked.

IMO

BC

[ondebanks]

Hi,

Sensor+ uses a special method of binning pixels that significantly reduces aliasing, so I would guess that the result will be significantly different from "fat pixels".

It is described here: https://www.captureintegration.com/wp-content/uploads/Phase-One-Sensorplus.pdf

The enclosed image is from the above document. Top represents traditional binning while bottom represents sensor-plus binning.

If you want the large pixel look, you could simply try downscaling a 60 mp image to 15 mp using nearest neighbour. (I actually know nearest neighbour is not the best emulation of large pixels, but it is a much better emulation than bicubic).

All this is oversimplification, of course. Large pixels will give more color aliasing. The suggested method of downsampling a demosaiced image would have much less colour aliasing than a large pixel image. Fill factor also plays a major role.

Best regards
Erik

With the exception on purely B&W sensors, one cannot really emulate "fat pixel" sensors by downscaling, such as converting a 60 mp image to 15 mp - even though ending up with 12 micron pixels does seem to be really "fat".

Firstly, one should downscale it in RAW form, before any de-mosaicing, because 4x [6 micron de-mosaiced RGB] pixels is nothing like a 1x [12 micron R or G or B] pixel.

Secondly, how best to combine the pixels in the RAW is problematic, and doesn't match a native "fat pixel". Traditional binning combines 4 pixels in a particular colour over a span of 18 micron blocks with internal gaps, not a solid 12 micron area. The Sensor+ algorithm is more elegant in the green channel, where the span is diagonal, non-overlapping, less gappy, and 17 microns on a side rather than 18; but Sensor+ still does exactly the same as traditional binning in the red and blue channels; overall it is far from being the same as a fat pixel back of the same resolution.

Because of these issues, only a fat-pixel colour back can give you that fat pixel look.

I also wonder how much of the "look" is due to the noise characteristics of the back. Fat pixels are characterised by "high signal, high noise". With newer CCD backs the trend is towards "middling signal, middling noise". Sensor+ is different again, with "high signal, middling noise". Some people apparently prefer the texture of images from the old backs, even if they are not aware of or cannot quite elucidate why that is (it's their "high signal, high noise" nature).

Ray

[bcooter]

does this mean  the only CMOS to make the (your) top was the original 1Ds?

No.

There is no going back.

From a professional standpoint I would never return to a 1ds for heavy production (though I kept one of my bodies).

Though if something is still viable like my p30+ or RED1's they continue to get use, because they're still useful and produce robust files, though less useful due to size and speed.

My industry is changing all the time and faster production, more natural settings, more spontaneity is the standard for today and fast cameras, slightly smaller form factors work well in those scenarios.

Last week shot a production using a tricked out 70d for motion, a 1dx for the stills and they were fast enough and produced a file good enough for the creative brief.

But if the project was studio based, high production with a lot of artificial lighting, it would have been back to larger cameras, (both in still and motion).

It's all about using the right equipment for the creative brief as there is no perfect camera for everything.   

Interesting that Wednesday, my black magic pocket camera came in.  I paid $499 and honestly would never have bought it at it's original price.

I really thought it was a camera I'd try, then probably sell a week later, but yesterday doing some quick testing, I'm enamored with the file and form factor.

Less so with the crazy use of 7 batteries a day and the inability to format a sd card in the camera,  but this camera will have an important place in my kit as in early quick tests the file grades very well.

The bmpc reminds of what would happen if leica made a cinema camera, as it's kind of a one lens, one look camera with it's own set of quirks.   

For our work if I really had to shoot everything with one kit, it would be the case of modern Canons which does everything well, just not overly spectacular, but the file isn't the goal, the creative brief is.

I just mentioned the 1ds because it's interesting that in 10 years or so, it produces a prettier file out of camera than more recent versions, just not enough to discount more modern cameras.

IMO

BC

[ErikKaffehr]

Ray,

Thanks for putting things "right" and giving a lead, on "phat pixels". So, you feel it is about noise characteristics?

Best regards
Erik

With the exception on purely B&W sensors, one cannot really emulate "fat pixel" sensors by downscaling, such as converting a 60 mp image to 15 mp - even though ending up with 12 micron pixels does seem to be really "fat".

Firstly, one should downscale it in RAW form, before any de-mosaicing, because 4x [6 micron de-mosaiced RGB] pixels is nothing like a 1x [12 micron R or G or B] pixel.

Secondly, how best to combine the pixels in the RAW is problematic, and doesn't match a native "fat pixel". Traditional binning combines 4 pixels in a particular colour over a span of 18 micron blocks with internal gaps, not a solid 12 micron area. The Sensor+ algorithm is more elegant in the green channel, where the span is diagonal, non-overlapping, less gappy, and 17 microns on a side rather than 18; but Sensor+ still does exactly the same as traditional binning in the red and blue channels; overall it is far from being the same as a fat pixel back of the same resolution.

Because of these issues, only a fat-pixel colour back can give you that fat pixel look.

I also wonder how much of the "look" is due to the noise characteristics of the back. Fat pixels are characterised by "high signal, high noise". With newer CCD backs the trend is towards "middling signal, middling noise". Sensor+ is different again, with "high signal, middling noise". Some people apparently prefer the texture of images from the old backs, even if they are not aware of or cannot quite elucidate why that is (it's their "high signal, high noise" nature).

Ray

[PWS]

thx a lot to all of you guys...
for me the conclusion seems to be that there is no 'magic'
as ray explained it's more about the back's noise.

some may call it 'magic', others just don't see any difference...

now there's one more thing that just came up my mind.
erik was talking about downsizing-/sampling methods.

i'm using photoshop cc currently and just noticed that it has 'automatic' method for resizing pictures.
i was wondering for weeks why my jpg's look too sharp after resizing them to web size.
now since i switched to 'bicubic smooth gradients' the results look much better.

so what's the best method for resampling pictures?

[eronald]

No.

There is no going back.

From a professional standpoint I would never return to a 1ds for heavy production (though I kept one of my bodies).

For our work if I really had to shoot everything with one kit, it would be the case of modern Canons which does everything well, just not overly spectacular, but the file isn't the goal, the creative brief is.

--snip--

I just mentioned the 1ds because it's interesting that in 10 years or so, it produces a prettier file out of camera than more recent versions, just not enough to discount more modern cameras.

IMO

BC

What would you keep if you had time, and just wanted a good file?

I don't see any reason a "heavy production" camera should be expected to have an especially good file, just like a sports camera ... other attributes become the priority eg. tethering or wireless, focusing, color matching of the file, frame rate  etc.

The 1Ds was incapable of doing a fashion show without hitting the buffer ...with the cards of that day.

Edmund

[ondebanks]

Ray,

Thanks for putting things "right" and giving a lead, on "phat pixels". So, you feel it is about noise characteristics?

Best regards
Erik

Hi Erik,

Well when people talk about textural things, such as "film-like grain", then yes, I put that down to the noise characteristics.

Of course there are other typical "fat pixel" characteristics as well - the pleasing colour from Kodak's and Dalsa's choice of CFA bandpasses, the lack of an AA filter, the high lens MTF per pixel...these are all elements in "the fat pixel look".
I'd sum it up as The Four C's: Crisp, Colourful, Contrasty, and Crunchy noise. 

Ray

[ondebanks]

(The kodak dcs 760 files were very pretty, but at 6mpx lacked the detail needed for this project, but had Kodak continued with this camera to around 11 to 18mpx I'd probably still be using it).

BC mentioned the APS-H 6MP Kodak DCS 760 above - this had almost the same 9 micron Kodak sensor tech as all the 16 MP square backs and about half of the 22 MP backs; it had a less efficient CFA, counterbalanced by lower readout noise; but it had the same pixel size, same relationship with lens MTF, and same lack of AA filtering (the filter was an optional item placed ahead of the mirror, like in the Sigma Foveon DSLRs). So whatever "fat pixel" magic those MFDBs had, it should have had as well...just less enlargeable.

BC, does that tally with your recollection of its image quality?

Ray

[Theodoros]

You could give the commercial names of the ones you talk about in your post ...
Or say the magic words that summon the appropriate rep and make him reminisce

Edmund

I've used 37x49mm only out of the Kodak 9mμ sensors, but I am experienced with almost all of them... P1 P25+ is a great back, I would prefer it over P45+ any day... Imacon 528c is what I use, the Imacon 132 should be the same but without the MS capability, it seems to be slightly worst though... it seems that they where hand selecting the best sensors to fit in the MS backs... Sinar 54HR has the disadvantage of being a tethered only back but in studio, it is as good as the 528c... The Sinar 54m beats them all IMO in single shot mode, if it wasn't tethered only it would be my first choice for single shot... Out of Dalsa sensor, I used to own Sinar Emotion 22 some years back... Colour accuracy was much better than the Kodaks, it was very close to the 75lv but not quite there... but other than the colour, the back was nowhere near the Kodaks, it seemed to luck the overall "bite" the Kodaks have... The Mamiya ZD I've only tried on Mamyia AFD... The ZD is the worst back I've ever tried IMO, Aptus 22 seemed to be a copy of Emotion 22, but with slightly worst colour accuracy than the Sinar (obviously due to software) but still more accurate than the Kodaks... Aptus 5ii shared the same a little warmer cast that Aptus 22 also had, but it was sharper and more dynamic, much better at ISO 200 than the other Dalsas too... If I was to go with Dalsa, I would aim for the 33mp sensor than the 22mp one... In fact I love the Dalsa 33mp sensor as much as I hate the 22mp one... It's the other way around with the Kodaks though... I much prefer the 22mp sensor than the 39mp one with Kodak... As for the unique "fat pixel magic" special look... as I said, it only exists with the Kodak 9mμ sensor IMO...