Measuring rolling shutter: put a number on this issue!

Yes, it would be nice if we could measure all the modes, as we did with the GH4 thanks to blazer003.

My guess would be that 1080p60 FF is skipping half the lines, whereas 1080p60 APS-C is reading everything and doing a bad downsample, and 720p120 APS-C reads half the lines and then does a bad downsample.
This would mean:
a7s FF 1080p24: 30.5 ms
a7s FF 1080p60: 15.8 ms
a7s APS-C 1080p24 and 1080p60: 19.5 ms
a7s APS-C 720p120: 9.8 ms

So far this is just a guess, though
 
For anyone still wondering about the accuracy of my measurements: they seem to be just fine. I just found this blog where they measure the sensor readout speed of the a7s using some fancy equipment, and the numbers are close enough:

a7s FF: I said 30.5ms, they say 34ms
a7s FF: I said 19.5ms, they say "a little less than 20ms"

(his numbers may be at 3:2, so then a 16:9 crop would be 28.7ms and 16.9ms)
 
Last edited:
eddavid said:
Do the NX1 Samuel - I'll shoot a test!
Click to expand...
Yes, please! I'm very curious to find the RS figures for this camera. Samuel has the procedure on the first post. I'm sure that if you made the required shots and didn't want to make the calculations yourself, he or someone else would be willing to do it.
 
I still have no proper test for the NX1, a7II, or 7DII. I'd love to get them, particularly for the NX1. I have no interest on the 7DII, and same for the a7II unless the IBIS is useful for video (which seemed not to be the case with the original firmware, but may be with the recent update).
 
Thanks!!

I ran my own measurements over that same clip and came out real close to your number: 30.6, 31.6 and 31.4ms, for an average of 31.2ms in 4K mode.

That puts the NX1 at the bottom of the table, but basically tied with my a7S in FF mode.
 
I'll submit a 1080p & 4K test with the Nx1 shortly. In 1080p there is very little rolling shutter and I'm curious how it will place on the chart.
 
Thanks, mate!!

These roughly confirmed the measures we already had for 4K and UHD (30.9ms and 32.6ms), and also that the 1080p mode is a lot faster. Like A LOT faster. It's actually the fastest camera in the chart, and by a very wide margin: 7.9ms. Amazing.

I think right now, if you can't get an FS7, the NX1 is the camera to get for green screen, and also for fast action shots. If they add a gamma mode that matches the DR of my a7S+slog2, it's the all-round winner.
 
Well, the GH4 has two weak points: it has a lot of noise (so not particularly good for green screen) and relatively little DR (so, not good in general if you care a lot about DR, as I do). So many great options these days, each one gets to buy the camera that best fits their taste/needs.
 
Using both the NX1 and GH4, i find the DR of the GH4 slightly better (principally on the highlights), but it's just my opinion. The Gh4 has a lot more noise yes but i think the Noise Reduction of the NX1 is too strong, too bad we cannot have control on it.
 
Last edited:
I can confirm improved rolling shutter in 4K on the NX1 with the 1.2.1 update. I will shoot a quick test and email it to you samuel!

Also, really annoying having every thing in the camera reset when you update it.
 
Last edited:
Thanks again.

I ran the numbers, and I don't see a significant difference. One could want to see a difference, but a statistician would complain

Original measurements:
NX1 1080p: 7.9 ms (7.7-8.0-8.1-7.8)
NX1 4K: 30.9ms (30.6-31.6-31.4-30.7-30.2)
NX1 UHD: 32.6ms (32.9-32.0-32.9-32.5)

New measurements:
NX1 1080p: 8.2 ms (7.9-8.4)
NX1 4K: 30.1ms (29.6-30.6)
NX1 UHD: 32.2ms (31.6-32.8)

Fitting this data to a model to have a bit better power for inference (at the cost of possible misspecification):
My model says UHD rolling shutter is 4 times as much as 1080p rolling shutter, and 4K rolling shutter is 3840/4096 times the UHD rolling shutter; using that, I transform all my measurements into "UHD-equivalent" measurements, and compare the old and new numbers (I have a lot more observations to compare thanks to these assumptions). 1 is the original firmware, 2 is the new one, and the conclusion is that measurement error is bigger than any improvement that the new firmware may have brought:
nx1.jpg

The only optimistic observation here is that the two fastest measurements from the original batch come from the 1080p mode, which is the one with most relative measurement error (and in which the "model specification" is most likely to be incorrect). If you take those two away (or all the 1080p measurements), then you'd have three (two) values in the new batch that are outside the range of the old batch, and it's possible a statistic test would say the new firmware has faster RS. Pointless, I won't run the numbers, measurement error is there and is too big to allow identifying small improvements, if any.
 
Last edited:
Both the D750 and D810 offer great video from an awesome stills camera, I'd like to add them to the table if somebody shot the test...
 
You must log in or register to reply here.
Back
Top