The mystery of the Sony A7 III vertical stretch - was it a hotfix for blinking pixels?

[Andrew Reid]

Although we shouldn't worry too much about this because it's barely noticeable, it's an interesting mystery to test and solve for Sony.

Read the full article

[Robert Collins]

Hmm... As the fix for this is to expand the X axis in post 1% or presumably decrease the y axis by 0.7% or something (maths not my strong point). I would conclude.

Option1... hot fix for dodgy horizantal lines. A 1% adjust wont wont fix that.

Option 2... Digital lens correction. Well then it would appear in both 24p and 30p

Option 3... FCPX bug. Maybe...

Option 4? Sony is down rezzing the video from say 5 or 6k to 4k and screwed up its algorithm. Quite probable.

[Andrew Reid]

30p is a crop, so the lens correction wouldn't reveal the very bottom edge of the sensor readout.

Looking at Caleb's video, the stretch is certainly enough to push the bottom of the frame out of the picture where the blinking pixels occurred.

My theory is that Sony discovered some units shipped with a hardware fault causing corruption on that bottom scan line of the readout, and only way to fix it was to hide it with firmware fix, rather than do a recall.

Why else would it be stretching the picture? It has no purpose and is an unlikely "bug" (would be the first bug of its kind ever in the camera industry). It is a full pixel readout in 4K/24p/25p of the entire sensor top to bottom, so yes they are scaling from 5 or 6K down to 4K. I don't think that's the issue. They have done it before, no problem. Wonder if A9 has same stretch?

[MdB]

The 30p mode has a slight crop in full frame mode, which would probably explain why it doesn’t occur then (by whichever means it does occur). It’s unlikely to be an FCP thing as video doesn’t have any ‘spare’ realestate like stills do. What you see is what you get. But either the other two might be right. I mean there are all sorts of funny artefacts at the edges of sensors (things like debayering gets mixed p because there aren’t the surrounding pixels etc), so if the camera was fully sub sampling right to the edge that would likely cause visible issues. These won’t occur in the 1.1x or 1.5x crop modes, but could occur at full read. It’s also likely to ‘flash’ and cause other anomalies as the the sensor is constantly combining the read out pixels, those ‘duds’ will be mixing with perfectly normal pixels and flicker between the two. A bit like when you have a stuck pixel on a sensor and it flickers on and off on the LCD when not in 1:1 magnification mode.

Any sensor would / could do this. They normally have a bit of run off of unused pixels around the edges. But the downsampling is where it becomes highlighted. 

[Robert Collins]

1 hour ago, Andrew Reid said:

30p is a crop, so the lens correction wouldn't reveal the very bottom edge of the sensor readout.

Looking at Caleb's video, the stretch is certainly enough to push the bottom of the frame out of the picture where the blinking pixels occurred.

My theory is that Sony discovered some units shipped with a hardware fault causing corruption on that bottom scan line of the readout, and only way to fix it was to hide it with firmware fix, rather than do a recall.

Why else would it be stretching the picture? It has no purpose and is an unlikely "bug" (would be the first bug of its kind ever in the camera industry). It is a full pixel readout in 4K/24p/25p of the entire sensor top to bottom, so yes they are scaling from 5 or 6K down to 4K. I don't think that's the issue. They have done it before, no problem. Wonder if A9 has same stretch?

Still if they fixed the 'scan line' corruption with a 'y axis stretch' of 1% you would think they would be smart enough to also stretch the 'x axis 1%' too.

[ND64]

I was thinking about the complexity of our visual system that recognizes such small differentiations without any chart.  

[jonpais]

???

[mojo43]

I am wondering if this is even a problem? I guess it could be... Tons of lenses have distortion which is very obvious. This problem to my eye completely unnoticeable. Does it matter?

[Andrew Reid]

No it doesn't matter.

But it is interesting to know why it happens, as it's very strange.

[kye]

My impression was that it was a quick 'fix' to bypass some kind of issue.

Considering how important aspect ratios are in both still and moving images, and the fact that everyone has known they're important for almost the entire history of photography, it's unlikely it was a mistake.

Of course, regardless of how and why it happened, it's likely they'll fix it quietly and we'll never find out

[Andrew Reid]

The aspect ratio hasn't changed though.

Distortion has.

[Mark Romero 2]

On the other hand, it's great for filming short, stocky talent...

[kye]

11 hours ago, Andrew Reid said:

The aspect ratio hasn't changed though.

Distortion has.

You're right about the size / resolution of the output files not changing, but the aspect ratio of the things being filmed has changed (circles aren't round any more).

[jonpais]

5 hours ago, kye said:

You're right about the size / resolution of the output files not changing, but the aspect ratio of the things being filmed has changed (circles aren't round any more).

My videos look anamorphic since the damn update.

[padam]

From some earlier comparisons, I think the A9 does the same thing, so I think it is persistent in FF 6k readout mode - and as usual the answer for the fix will be called A7SIII.

[jonpais]

39 minutes ago, padam said:

From some earlier comparisons, I think the A9 does the same thing, so I think it is persistent in FF 6k readout mode - and as usual the answer for the fix will be called A7SIII.

Link? Can you corroborate that statement?

[padam]

You can find A9 video comparisons on YT, but here is one:

You can see the exact same stretching.

 

So in conclusion, this is how the computation works in the current FF cameras capable of 6k readout and it is not going to be "fixed'.

I'm pretty sure that the A7SIII will have less megapixels so the internal downsampling won't be pushed to its limits. But for 4k60p they may need to upgrade the codec and processing to be able to handle that (maybe with an external recorder, but probably 8-bit 4:2:2 limited).

[jonpais]

13 minutes ago, padam said:

You can find A9 video comparisons on YT, but here is one:

You can see the exact same stretching.

 

So in conclusion, this is how the computation works in the current FF cameras capable of 6k readout and it is not going to be "fixed'.

I'm pretty sure that the A7SIII will have less megapixels so the internal downsampling won't be pushed to its limits. But for 4k60p they may need to upgrade the codec and processing to be able to handle that (maybe with an external recorder, but probably 8-bit 4:2:2 limited).

There is no stretch in the a9.

[padam]

There is, just look at the houses on the right before punching in. If it was not rotated it would be more obvious.

Also visible here:

 

[Papiskokuji]

I’m amazed at how only few people are concerned by this image streching issue ! As everybody’s always so eager to attack every manufacturer for anything. I find this ratio problem much more worrying and shocking than even overheating. Eventhough there’s lens distorsion coming into play when it comes to proportion, I find it puzzling that people let Sony go with this one... I can’t imagine myself explaining to a client why its product looks streched on this or this packshot. For me it’s the basics : having a 1:1 aspect ratio...

Hope it’s gonna be fixed later with a new firmware but I need more people to complain about it

I’m waiting for the A7S III anouncement before I buy anything but I seriously consider buying the A7 III, if and only if this issue get fixed (not that anyone cares!)

Am I the only one shocked by the stretching thing ?