Backlight landscape photography realized - say no to silhouette

[voidshatter]

This thread is all about pixel peeping for high contrast landscape shots. It is not about studio shots. I do not mean to bash any existing exotic gear, but I simply cannot resist how the latest technology evolves.

Ansel Adams never deliberately challenged the sun. He developed the "Zone System" to optimize exposure for dynamic range and pixel peeping. While he emphasized the importance of the film, I bet the sensor (or digital back) also plays an important part, at least as important as the glasses!

Here is the Dante's Inferno I have just visited: Phase One IQ250, Alpa 12 SWA, Rodenstock 23mm HR, Rodenstock 40mm HR

Below is my test shot at Durdle Door, UK. -6mm shift downwards was applied to the Rodenstock 23mm HR lens, with Rodenstock centerfilter (2.5 stops), Lee ND grad 0.6 (2 stops) and Lee Big Stopper (10 stops).

https://500px.com/photo/92736199/

This was a single exposure (no blending or bracketing) to stress test the dynamic range of the Sony CMOS sensor. The shadow recovery is as awesome as that of the Nikon D800 (similar Sony sensor technology)!

 

[voidshatter]

Gearhead is all about taking pictures of what the others cannot easily take, and enjoying pixel peeping. I like shooting directly into the sun, not just because the light condition looks romantic and gives a visual impact, but also because it is a great challenge for the dynamic range of the sensor and the flare resistance of the lens. It is as exciting as exploring the uncharted lands!

While the others were shooting star trails, I shot the sun trail: https://500px.com/photo/67886275/

I was able to capture the trail of the sun by stacking a 77mm B+W 10-stop filter, a Lee Big Stopper and a Lee 0.6 ND Grad Hard.

This is a single exposure of over 2 hours. Yes, it is processed from a single RAW file and there is absolutely no other external materials used from other exposures. It is not a stacking of multiple frames. No dark frame subtraction was performed to reduce the noise, and the battery life did not permit long exposure NR either. The sensor of the D800E is amazing in maintaining a control of hot pixels throughout the ultra long exposure. The sensor of the D4S looked so much inferior in terms of noise control, even for merely 45 minutes.

 

[voidshatter]

I decided to enter the Dante's Inferno after I roughly compared a D800 (at ISO 100) against my friend's Phase One IQ260 (at ISO 50). Thanks to the anti-aliasing filter and the inferior glass of the D800, the details were totally overwhelmed by the Rodenstock lens. Yes there are lots of talks about larger sensor, 3D-look, etc. The advertisement says 1-hour long exposure capability, which seemed perfect for my kind of tastes. Okey, it is also the largest sensor (even slightly larger than the IQ280!).

However the adventure was not full of excitement but also mixed feelings. I don't mean to bash the larger CCD sensors - they are great products. Perhaps these are just not fit for my purposes.

The first thing I noticed about the Dalsa 60 MP sensor was the lithography partitions, which causes tiling issues:

I noticed the vertical splits of the image where brightness differs slightly of the partitions. Just to rule out that I had a poor sample, I downloaded the official sample images from Phase One, and the other IQ260 shots also had the same issue when I tried to pull local micro-contrast heavily. So, this is a general issue for the larger Dalsa CCD sensors. An LCC shot cures it most of the time, but sometimes the problem is still there, especially true for B&W.

 

[voidshatter]

The second thing I noticed is the poor Signal-to-Noise Ratio (SNR) I get from long exposure shots of the IQ260. At ISO 50, yes the IQ260 produces clean images, with decent shadow recoverability. However, for long exposure shots, it is required to use a dedicated "Long Exposure Mode", for which the lowest possible setting is ISO 140! This is a nightmare for dynamic range, as it severely impacts the SNR in the shadow! When I compared it against a Nikon D800E side by side I was totally shocked:

 

[voidshatter]

I was depressed and then I headed to compare other digital backs against the D800E. In terms of shadow recoverability, only the 44x33mm Sony CMOS sensor (e.g. IQ250) could match the D800E! The IQ280 simply has no shadow recoverability when long exposure builds up the heat. (Don't even think that I am telling lies. I have uploaded RAW files for you to study: https://www.dropbox.com/sh/imzk9yp3uudv63c/AACSxr3vcN3kSRQQ1vMQjX9Ka?dl=0)

Just to rule out that I have poor samples of CCD sensors, I also downloaded RAW files from http://www.getdpi.com/forum/medium-...dt-tech-cam-test-iq250-vs-iq260-vs-iq280.html and also obtained the same conclusion that the IQ250 has obviously more usable dynamic range:

 

[voidshatter]

To show how this SNR in shadow (usable dynamic range) affects real world pixel peeping, I have the following examples.

Here is an IQ260 shot with a Lee ND grad 0.9 (3 stops) and a Lee Big Stopper (10 stops). As you could see from the histogram, I employed "Exposure to the Right" (ETTR) and the exposure was almost perfect. The highlight was almost blown, so that the SNR in shadow was optimized.

Here is another shot with the Nikon D800E with No ND grad at all! It was aggressively underexposed to preserve the highlight details.

Here is a comparison between what you can get after processing in Capture One:

As one can clearly see, noise reduction would only hurt details. The low SNR in shadow (less usable dynamic range) for the IQ260 results in less shadow details for pixel peeping! The D800E outperforms, simply due to the superior sensor technology in dynamic range, even with inferior glass!

 

[voidshatter]

With clear conclusions that the 645 CCD sensors can no longer match the Sony CMOS sensors in terms of dynamic range, especially in the long exposure territory, I then gave up all the fun to play with single exposure, and tried to do blending with exposure bracketing. I had no luck either (for pixel peeping).

First of all, it is clear that the long exposure of the CCD sensor would rely on the darkframe noise reduction (also known as "black calibration" or "long exposure NR"). It works in the way that for each long exposure of X seconds, another exposure of X seconds with the shutter closed is taken, and any incremental hot pixel is then subtracted from the original exposure and removed. This is a disaster for sunrise and sunset, as the light condition changes rapidly, and while you wait for the darkframe you would have to always make a hard decision whether to give up the current frame by pulling out the battery to get the next shot for the decisive moment, or let the previous frame finish and miss the current beautiful light you have.

For a long exposure sunset scenario, due to the limitation of the darkframe noise reduction, exposure bracketing can only be taken before the sunset without the ND filter to expose for the foreground (so that the darkframe does not occupy the beautiful moments). For very long exposure there is basically only window for one shot to capture the sky. This causes a problem that by taking off and putting on the ND filter, the angle of view of the lens changes slightly, and the very minor distortion introduced by the ND filter is non-linear. It is then a pain to try to align these two exposures accurately and perfectly in Photoshop. Yes for unlimited time and effort it is possible to align these layers close to perfection, but it is just too much harassment for me. Neither Photoshop nor Helicon Focus could do it automatically for me. I would have to do free transformation manually. A misalignment would cause drop of sharpness in the transition area. Also note that due to the nature of the digital pixels, any free transformation (e.g. distortion correction in post-processing) would result in a decrease in sharpness.

 

[voidshatter]

After all, the Sony CMOS sensor is now in the top league for my applications, and the dxomark thing is telling the truth for me.

The sensor in the D800, D800E, D810, A7R are all based on the Sony IMX094 chip, which is really a gem. Below is a picture of sun trail taken by a friend of mine, which is a single insane long exposure of 3 hours! (D810)

 

[voidshatter]

When the light is right for the CCD sensor, the Rodenstock HR lenses really outperform the 35mm format lenses. It is such a shame that the CCD sensor is now the limiting factor for the final output of a landscape picture (high contrast scene).

To resurrect the dignity of the medium format in the long exposure territory, and renown what the Phase One P45+ used to bring us (that it used to be superior than the Canon and Nikon), I have finally made the decision to switch to the smaller sensor, the IQ250 instead. Yes, it is only 44x33mm in size, which is merely larger than the 35mm format sensors. People snorts that medium format digital starts at 49x37mm. I chose to ignore these and headed to the smaller sensor, because it is a Sony CMOS sensor!

The first advantage about it is that you could immediately switch off the darkframe noise reduction by choosing "Aerial" in "Camera mode", and "Zero" in "Shutter latency". With these settings, each time you boot the digital back, starting from the second shot, you do not receive the darkframe NR count down. By disabling this you still get hot pixels but these are quite fixed in pattern and can be eliminated by sliding the "Single pixel" bar to "1" in "Noise reduction" in Capture One.

With the CCD sensors you could of course do the same to disable the darkframe NR, but the images would be nowhere as usable as what you can get from the Sony CMOS sensor!

 

[voidshatter]

I went back with the IQ250 to test the same scene where the IQ260 was bashed hard by the D800E. I deliberately agressively underexposed the scene just to protect as much highlight details I can. Then I applied the following brutal adjustments in post-processing:

+4 exposure
-100 highlight
+100 shadow

Finally, with the help of the tilt-swing function (Scheimpflug principle) on the technical camera, and the really superior glass offered by Rodenstock HR series (in this case the 40mm HR), I am now able to defeat what I can get from the legendary Sigma 35mm f1.4 Art lens on the Nikon D800E! Here are the RAW files for you to do pixel peeping, so that you know I am not telling lies: https://www.dropbox.com/sh/hhgursjx0u44j6g/AAD11DFbEGsqL6TsKWCGPz0Da?dl=0

The Rodenstock 40mm HR lens was tilted by about 1 degree and focused at about 5 meters away. The Sigma 35mm f1.4 Art lens was focused at the bicycle. Each lens was shot at f/11.

As all can now clearly see, when the advantage of the technical camera (ultra high resolution glass) is combined with the Sony CMOS sensor (ultra high SNR and clean shadow), it is simply beyond what you can achieve with the 35mm format!

 

[voidshatter]

One more test shot to demonstrate the pixel peeping for night scenes (long exposure + high contrast):

I haven't optimized the shimming and HPF ring yet for this combo, and it is nothing short of jaw-dropping already. Finally I have a taste of the superiority of the medium format (with a very small sensor )!

 

[voidshatter]

My personal advise for high contrast landscape shots:

It is probably best to wait 1-2 months until the rumored Sony A9 (46 MP - 54 MP in 35mm format) is announced. Then the prices of the current digital backs around that MP count would crash. If you have to get a high dynamic range medium format digital back now, then it is perhaps best to get the Sony CMOS sensor (IQ250, IQ150, Credo 50, H5D-50C, H5D-200CMS, CFV-50C, 645Z). Personally I would recommend the Hasselblad CFV-50C for the price being only around $10K in Asian countries (e.g. Japan, China etc). It offers live view and can be used on a technical camera.

What to expect in the near future:

a) According to a trusted source, Phase One already has a 120 MP CMOS sensor ready (in 54x40mm fullframe 645 format ). It is likely to be the same pixel density as that of the Nikon D7100 (also the rumored upcoming Sony A9). They choose not to announce it now as there is virtually no wide angle lens compatible with it (i.e. crosstalk-free and high-resolution), and they do not want to impact the sales of the current products. Note that Phase One has ended their Investment Protection Plan in September 2014 so try to estimate when they will announce the new killer digital back. If that new back becomes reality, then newly designed extremely retro-focus wide angles would be required. Perhaps Rodenstock will design new lenses in the Digaron SW (yellow-banded) range? I expect it to be even heavier than the current 32mm HR and it would perhaps employ Copal 1 shutter (or even just rely on the Alpa FPS platform?).

b) Whether you like it or not, CCD has had its gold ages and eventually it will be replaced by CMOS. Leica now lists S Typ 007 as their new flagship, which means that they regard CMOS more advanced than CCD. It is interesting to see that Hasselblad (or B&H) now regard CMOS to be superior as well (i.e. a smaller CMOS is even priced higher than a larger CCD):

If you really prefer CCD, then perhaps the H5D-40 is a good choice as it is now half-priced:

Expect the same price crashs on the IQ250, IQ260 etc when the 35mm format sensor advances in the upcoming new year.

 

[voidshatter]

One more example to show what to get from the IQ250 + 23HR combo (+6mm shift) You can't beat this now with IQ260 or IQ280 for such a 5-minute long exposure like this. This is a Sony CMOS sensor! ND grad is a no go, as it would cut the tree and if you attempt to recover details of the tree you would still rely on the dynamic range of the sensor. Trust me, for an IQ260 or IQ280 even the cloud above the sun would become noisy!

 

[ondebanks]

This is all very useful - thanks for sharing!

One comment I'd make, especially regarding some of your earlier image comparisons, is how essential it is to give equal exposure times (for a fair dark noise comparison) and equal exposure to light (for a fair shadow/shot noise comparison). This means that you should set the same shutter speed and f-stop, even if the ISO settings are different on the cameras.

Now what I am saying may sound wrong - we are all trained photographically to automatically compensate a change/difference in ISO by a balancing change in aperture and/or shutter speed. And that's what our light meters do. But not in this testing scenario. So if your DB is best at ISO 50 and your DSLR at ISO 100, don't halve the DSLR exposure time with respect to the DB exposure time to make them "equal". Shoot with the same time and f-stop (and filters), to keep the test as "apples with apples".

Ray

 

[ondebanks]

Also, this caught my eye:

With the CCD sensors you could of course do the same to disable the darkframe NR,

So it is possible to disable the darkframe in a Phase One CCD back? And that's just in the IQ series, not the P+, right?

Ray

 

[voidshatter]

This is all very useful - thanks for sharing!

One comment I'd make, especially regarding some of your earlier image comparisons, is how essential it is to give equal exposure times (for a fair dark noise comparison) and equal exposure to light (for a fair shadow/shot noise comparison). This means that you should set the same shutter speed and f-stop, even if the ISO settings are different on the cameras.

Now what I am saying may sound wrong - we are all trained photographically to automatically compensate a change/difference in ISO by a balancing change in aperture and/or shutter speed. And that's what our light meters do. But not in this testing scenario. So if your DB is best at ISO 50 and your DSLR at ISO 100, don't halve the DSLR exposure time with respect to the DB exposure time to make them "equal". Shoot with the same time and f-stop (and filters), to keep the test as "apples with apples".

Ray

We had a debate about the "native ISO" a while ago, and we couldn't agree with each other

I think the best thing to do is to look for the blown highlight in the whole picture, which can be done by Raw Digger. If you hover your mouse to a certain pixel of the picture, you get readings of the four channels (RED, GREEN, BLUE, GREEN_2). If you see that the level is approaching the limit (e.g. 65535 for 16-bit RAW file like the IQ250/IQ260, or 16383 for 14-bit RAW file like the D800E), then that means that area is blown and it is not possible to recover the highlight details there.

Any sensor is limited to this, and a useful image would be to prevent any interesting area from getting blown out. Then you aim for ETTR to optimize shadow SNR. This is how you get full use of the dynamic range of the sensor.

In this way, there is no need to worry about the ISO, aperture, shutter speed etc. You just aim to retain the same amount of highlight details, then compare the shadow SNR.

For example, in the above comparison we could see that the highlight around the light on the 2nd floor (1st floor in UK standard) is almost blown when you do pixel peeping with Raw Digger. That means they now have the same (or similar) highlight recoverability. Anything pixel captured brighter than that in the RAW file cannot be recovered. Then it is now a fair comparison on the shadow SNR, regardless of the aperture, ISO, shutter speed etc.

 

[voidshatter]

to keep the test as "apples with apples".

Ray

Again, for the very first comparison between the IQ260 and the D800E, it is also fair, when you look into the amount of highlight details recoverable. For the same area on the pillar with that reflection, they are both almost blown out, i.e. almost 16383 for the 14bit D800E and almost 65535 for the 16bit IQ260. If you overexpose any of them, then they have no ability to recover the details in the area on the pillar. Then it is a fair comparison in the shadow area, regardless of aperture, ISO, native ISO, shutter speed etc.

Keep in mind that the 23mm HR had a center filter on, so the IQ260 actually had an advantage in terms of control of vignetting. It was still beat hard. There is no doubt that the CCD sensors are no match against the Sony CMOS sensors. Have it a try yourself if you can!

 

[voidshatter]

Also, this caught my eye:



So it is possible to disable the darkframe in a Phase One CCD back? And that's just in the IQ series, not the P+, right?

Ray

I haven't had a chance to try it out on the P+ series. As far as I am aware of, this trick works on the IQ250 and the IQ260. I believe I have also seen reports that it works on the IQ160 but I am not sure if I remember correctly.

 

[mbn]

Thank you for taking the time to put this together.:thumbup:
Pretty impressive stuff!

 

[Uaiomex]

Wow voidshatter, you alone made the most interesting and entertaiining Getdpi thread of the year. Thanks for the xmas gift!
Eduardo
P.S. I'm one of those boogers claiming that DMF starts at 48X36. ��