Raw Dynamic Range: Exposure Latitude

By Rishi Sanyal

Class-leading 15.3 EV dynamic range? Not exactly...

We've come to expect class-leading Raw dynamic range performance from Sony Exmor sensors, which translates to impressive ability to record a wide range of bright and dark real-world tones. This allows a photographer to choose an exposure that retains highlights - such as bright skies and clouds - while still maintaining relatively noise-free shadows. The higher the dynamic range of your sensor (inherently linked to the size and noise performance of the sensor), the more noise-free those shadows remain, sometimes despite drastic underexposures. That's exactly what we did below, and to the a7S' credit, it holds up quite well to this type of shooting for high contrast scenes:

That said, Sony went as far as claiming an 'unprecedented' 15.3 EV of Raw dynamic range for the a7S, which would put it squarely above any full-frame camera that we, or DXO, have tested. So how does this claim hold up to reality? Well, not so well...

Our test

Testing Raw dynamic range in the lab isn't easy - most setups include transmission wedges or separate light sources of varying brightness to measure the range of tones a camera can record faithfully - all the way from the brightest patch that isn't saturated (clipped), down to the darkest patch that isn't swamped in noise.

One issue with these tests - such as the ones DXO does - is that they simply give you a number, such as '14 EV', which unfortunately doesn't translate in to something visually very meaningful. Rather than repeat DXO's (rather reliable) tests, we've chosen to expose our typical studio scene in a manner that tells us about the dynamic range capabilities of a camera. By underexposing our scene by approximately 6 EV, and then adjusting the exposure 6 EV in ACR, we can get an idea of the recoverability of shadow detail. This is inherently linked to the Raw dynamic range of a camera.

Why? Why is the ability to recover shadows linked to the dynamic range of the camera?

When you're trying to shoot a scene with a lot of dynamic range, you expose so as to not let bright tones clip to white. This often throws a lot of tones (typically in the foreground, if you're shooting landscapes) into the shadows, especially when viewed on the dim, low dynamic range displays of today. It's here in the shadows that tones tend to be swamped in noise when a camera has low dynamic range. The higher the dynamic range of the camera - be it because of its larger sensor, or lower read noise, or some combination of the two - the cleaner these shadows are. Which translates to a better ability to lift those shadows if you wish to make them visible. Naturally, then, the higher the dynamic range, the better the ability to recover underexposed photos, and, so, the higher the 'exposure latitude'.

To get an idea of Raw dynamic range of the a7S, especially compared to similar cameras, we look at the ability to recover shadow detail in our studio scene. Since our studio scene is typically well-lit (10 EV), there really aren't many shadows to look at, so we've modified the scene illumination and purposefully underexposed cameras by 6 EV. In the tests below, the studio scene is lit frm one side only, with approxmiately 7 EV illumination on a grey patch in the middle. Lighting from one side ensures more shadows on the left side (absent a light). Raw images were pushed 6 EV, with some additional shadow lifting to make darker tones visible. All images were brigthness matched across a number of tones to ensure similar brightness levels for all tones.

Lit from one side only, as opposed to our typical 10 EV illumination from both sides. Exposures were 6EV below our standard Raw exposure, then pushed 6EV in Adobe Camera Raw. This illustrates the latitude of each cameras' Raw files, which is also intimately linked to the camera's dynamic range.

Exposure Latitude

Our exposure latitude test, shown below, looks at the ability to recover severely underexposed tones - something you might want to do if you've exposed a high dynamic range scene for the highlights, or simply made a mistake. Cameras with more dynamic range show cleaner results, while cameras with lower dynamic range have noisier results, often due to a smaller sensor and/or electronics with a higher noise floor.

As explained in the section 'Our test' above, we underexposed every camera by approximately 6 EV at ISO 100 from an 'ideal' exposure (for the center grey patch, illuminated at ~7 EV), then lifted the exposure 6 EV in post, with some additional shadow lifting to make deep dark tones visible. Have a look at how the a7S compares with respect to its peers, below:

While the a7S performs admirably, it's not got quite as much dynamic range - and therefore exposure latitude - as its sibling, the a7R, when results are normalized. This was a bit surprising to us, given Sony's claims of 15.3 EV dynamic range, which would put it at least a stop above the a7R, according to DXO's measurements. In fact, DXO's comparison to its siblings show the a7S as having approximately 1 EV lower DR than its a7 and a7R siblings.

Since the Nikons have similar performance to the a7R, the a7S also falls slightly behind the D750 and D810. However, it remains far ahead of Canon's offerings, where limited dynamic ranges at base ISO means very noisy shadows - likely due to high 'downstream' read noise levels between the sensor and the off-chip analog-to-digital conversion in Canon DSLRs.

Conclusion

While the Sony a7S does display impressive ability to lift shadows - and therefore formidable Raw dynamic range at base ISO - when compared to offerings from Canon, it's not quite up to its higher resolution sibling, the a7R, in normalized comparisons. This goes somewhat against the popular notion that less pixels = less noise. While this may hold true for very low light, high ISO applications, and certainly holds true for pixel-level analyses, it doesn't always extend to normalized comparisons. This is probably partly because manufacturers work hard to scale pixel performance with shrinking pixels - which is why high resolution sensors stay competitive in low-light shootouts. With particular respect to dynamic range, we guess that the larger pixels - and the larger charge they hold at saturation - limit the ability to accurately record both high, full-well signals as well as low, single-digit photoelectron counts at base ISO. This can manifest itself as a higher quantization error componenent of the higher downstream read noise values extrapolated for the a7S from DXO data.