A7s Test with Light Meter Readings

lyonfilms

Well-known member
The test information posted in the Vimeo page is directly below:



I (Randolph Sellars) shot this test with Lyon Films (see credits) to evaluate their new Sony A7s camera. Our primary goal was to evaluate the dynamic range and high ISO capabilities in low light situations. We chose challenging but common situations like high contrast day exterior and interior as well as extreme low light with practical sources. In the high ISO tests, we stopped down accordingly so that we could judge noise with a consistent exposure and see if there were any changes in dynamic range. Most of the footage is "raw" and uncorrected. Due to very busy work schedules, we did not have time to grade examples from every test. We simultaneously recorded 1080P to both SD cards and a Pix 240 external recorder (split screen) so we could see the benefits of a higher bit rate and color space recording - especially after grading. Sorry, we did not have access to a 4K recorder - another day in the future! Note: some of the color noise that you may see on the wall in the candle test is the result of compression and is not apparent in the original. This is especially true of the footage that was graded with Neat Video applied. The model's face was lit entirely by the two candles approximately 1 ft. away. The model's face in the interior day scene was lit with a 4' - 4 bank kinoflo. All light readings were taken with a reflective spot meter. With some of the dusk/night bridge shots, the focus on the model's face is slightly off - and not indicative of the camera's sharpness. I'm just not that great at pulling focus by myself without a follow focus unit. Lyon Films and I would like to thank Picture This camera rental (Portland, OR) for providing the Zeiss CP 2 cine primes used in the test. Additional thanks to our soon to be famous actress Paige (Sunshine Girl) McKenzie for being an entertaining model.
 
Vimeo has apps for viewing on smart phones at least I use one for the iPhone.

Great tests. Thanks for posting...
 
Last edited:
But without a link to copy / paste from you can't get the video from the Vimeo app... Will have to check back on a computer..
 
So, in Slog2 XAVC-S:
* in the candle test the lamp shade is seven stops brighter than the face shadows
* the lamp shade is fine at T/2.8, but it clips at T/2.1
* the face shadows are perfectly fine at T/5.6, noisy but bearable at T/8 (neat video makes them clean but soft), they fall apart at T/11

That's a usable DR of 10 to 12 stops, depending on your taste for noise and noise reduction software. For me it would be >11 stops. Pretty impressive.

Recording externally, you lose a bit less than a stop in the highlights, you get about a stop cleaner shadows (they are fine at T/8, they still break apart at T/11). If the highlight thing could be solved, that would be around one stop extra DR because of the cleaner shadows. Or just about the same as XAVCS+NeatVideo but with more detail in the shadows.

Cine4 gets the lamp shade clipped even at T/11, so it has at least 5 stops less DR than slog2.
edit: I have now seen with my own eyes that cine4 has just a bit less DR than slog2, so my guess is that they didn't recover superwhites in the cine4 test (slog2 records 32-235 or something like that, cine4 is full range, so it needs a 16-235 output levels filter to get highlights back)



Thanks a lot for the test, that was pretty useful.
 
Last edited:
lyonfilms, Thanks for this test.
Am I right to assume the pix240 footage was shot in Prores? if so, in post, did you set the gamma to "full range"? The Pix 240 footage looks 0-255 range as apposed to 16-235 to me (or the other way around), which represents the pix240 footage as more "contrasty"
If you take the pix240 footy into Resolve and set to "data levels' instead of "video levels", do you see a difference in blacks/whites?

Am I assuming the pix240 footy is not showing full range correctly? Even the ref frame in the vimeo video shows a major difference in contrast. look at the lamp at the top, its clipped on the Pix240. Look at the shadow areas, they're more crushed than the xavc-s.
 
It's not just the pix240: there are several samples around already that show that external recording has issues with range and changes contrast. My original guess was that the camera was exporting full-range with a high black point but the recorder thought it was 16-235, so it stretched it, clipping the highlights; seeing that the clip in the highlights is about a stop or less, I no longer think that's the reason, I now think it's something that Sony does to the HDMI-out signal, and which hopefully could be corrected with a new firmware.
 
Yeah I think it's the camera because the image also looks more crushed on an external monitor. I hope Atomos and Sony work it out because prores doesn't look great and XAVC-S does fall apart if pushed too hard with motion.
 
With the black floor point at around 30-32 and the white clip point at ~232 around 1/5 of the coding range is lost in s-log2 XAVC-S. Not exactly a clever decision by Sony considering the 8-bit coding space.
It's a small wonder that there are actually some cases where XAVC-S s-log2 is usable...
 
XAVC-S seems to be recording 30-255, so it's losing around 12% of the range on each channel, and 31% of the total color space (11.5M colors instead of 16.7M)
 
XAVC-S seems to be recording 30-255, so it's losing around 12% of the range on each channel, and 31% of the total color space (11.5M colors instead of 16.7M)

The other gammas definitely go noticeably below 30 if not all the way down to 0. Will check for the precise value later.
 
That's wacky. Why would Sony do that?
The same reason Canon did it with CineStyle: to match what higher end cameras do. But those higher-end cameras can spare a third of the color space without creating any issues, since they can shoot 10-bit or more...


cpc: yes, I know it's a slog2 thing, the other gammas don't have this high-black issue
 
The same reason Canon did it with CineStyle: to match what higher end cameras do. But those higher-end cameras can spare a third of the color space without creating any issues, since they can shoot 10-bit or more...


cpc: yes, I know it's a slog2 thing, the other gammas don't have this high-black issue

Yeah well, sorry to be the one bringing the bad news, but the white clip point in s-log2 xavc-s is 232 or so, not 255. :)


edit:
Also, curiously, the green floor is a few values below the red and blue floor, so the lowest blacks are actually purple. Not sure if this is white balance related yet.
 
Last edited:
XAVC-S seems to be recording 30-255, so it's losing around 12% of the range on each channel, and 31% of the total color space (11.5M colors instead of 16.7M)

It doesn't do that in all gamma modes though. Cine2 must be flagging at 235 and slog2 seems similar.
 
I believe it's just like my 5N: the file does contain data that goes up to 255, but the metadata that tells Premiere about the used range seems to be wrong (or Premiere is reading it wrong), so it things it only goes to 235 and clips everything above that. Add 16-235 output levels and you get back your highlights, you can now see those 235-to-255 values. This 16-235 output levels thing is definitely needed with XAVC-S a7s slog2 footage on CS5, so I guessed it reaches 255. I haven't looked at the actual values, though.

I'm not sure the zebras take this into account BTW.
 
I believe it's just like my 5N: the file does contain data that goes up to 255, but the metadata that tells Premiere about the used range seems to be wrong (or Premiere is reading it wrong), so it things it only goes to 235 and clips everything above that. Add 16-235 output levels and you get back your highlights, you can now see those 235-to-255 values. This 16-235 output levels thing is definitely needed with XAVC-S a7s slog2 footage on CS5, so I guessed it reaches 255. I haven't looked at the actual values, though.

I'm not sure the zebras take this into account BTW.

I don't leave to metadata to decide, everything is imported explicitly full range in Resolve.
The actual ranges are:
32-232 Slog2
18-255 Cine gammas (except cine2)
22-255 ITU709 gamma
(Add or take one at the low end.)

Changing black level might affect the last two low limits, I haven't checked this yet.
 
Mmmm... ok... god knows why Premiere clips the highlights then, but ok...

That means slog2 is throwing away 21.5% of each channel, and 51.5% of the total color space: just 8.1M colors instead of 16.7M. OUCH!!
If it ends up looking fine, I don't really care, but really... OUCH!!
 
Back
Top