NO rolling shutter in Super Slow Mo!

[Jim Michael]

Originally Posted by Alister Chapman

Here's some footage from my shoot last weekend. Note the almost total lack of rolling shutter effects on the propellors and rotor blades.

Royal International Air Tattoo 2012 in Super SlowMo! - YouTube

At :30 I see a little looking at a single frame and it looks like he's under power so I'd expect a rotational velocity of ca. 2500 rpm. At 4:00 I don't see any but it looks like he's on final so maybe 1000 rpm? The Osprey and helicopters have relatively slow rpm and I don't see any there. So even on the worst case not really noticeable. Nice.

[Alister Chapman]

No, that's not rolling shutter. The Airbus A400M has curved contra rotating propellor blades. The blades are curved anyway.

http://www.aviationnews.eu/blog/wp-c...8/06/a400m.jpg

[David Heath]

Originally Posted by Alister Chapman

I suspect there is some line skipping in SSM up to 240fps (every other line) which explains the aliasing increase, but as the sensor is quad HD, that doesn't result in a resolution drop. The need to line skip (if that is what is happening) may explain why the faster scan rate isn't available in normal modes.

I'm pretty sure that what is happening is that it is skipping on a 2:2 basis - reading two lines, skipping two, reading two etc. Coupled with that is the way the sensor is fundamentally read, which in this mode is effectively the same as the C300 does it - directly forming R,G,B values for a pixel from a 2x2 Bayer block. (So it needs two adjacent Bayer lines to form a single pixel.)

As a quad-HD sensor it has 2160 rows. If every line was read in this way it would give limiting vertical resolution of 2160/2=1080 - as happens with the C300. Only reading out 2 rows out of four obviously halves this to give a figure of 540, and an expectation that anything beyond this will alias, and likely in luminance only.

Lo and behold, look at Adam Wilts results for the FS700 at ProVideo Coalition.com: Camera Log by Adam Wilt | Founder | Pro Cameras, HDV Camera, HD Camera, Sony, Panasonic, JVC, RED, Video Camera Reviews (second chart, the 240fps one) The horizontal lines start to alias, then diverge to a null zone - which marks twice the Nyquist limit. That's at 1080 lpph, hence it follows that the fundamental limit of alias free resolution is 540.

The other interesting thing from that chart is the symmetry, horizontal resolution is the same as vertical (540lpph) at 240fps. That seems to indicate it's doing exactly the same horizontally as vertically - hence reading two columns, skipping two, reading two etc. Another way to look at it is that in a block of 16 photosites it's reading 4 (a 2x2 block) and ignoring the other 12.

(Worth also looking at the 480fps chart below. It's clear there that the symmetry has vanished, the horizontal res is as for 240fps, but the vertical is much lower. Judging by the 2xNyquist zone, my guess would be it's now reading two lines, skipping four, reading two etc.)

[Alister Chapman]

Who says the sensor is read in the same way as the C300. That's speculation, it might be read conventionally. Not saying your wrong, maybe your right, but I don't think we can say for sure. The zone plates should show equal H & V resolution if it's read like the C300 but there is something odd about the processing or sensor readout in the FS cameras that results in asymmetric resolution at HD in normal modes.

Your right about the skipping being line pairs, forgot it's using some form of CFA, whether it's read like the C300 or conventional Bayer it has to work with 2x2 pixel blocks.

[David Heath]

Originally Posted by Alister Chapman

Who says the sensor is read in the same way as the C300. That's speculation, it might be read conventionally.

If it's accepted that it's quad-HD, it would make absolutely no sense to read it conventionally (by which I'm assuming you mean to fully deBayer and downconvert) for a 1080 camera - C300 style readout etc will give full 1080 resolution with minimal processing and power consumption etc. Why do it the difficult way if you can get the results far easier?

That said, all my previous post was referrng to 240fps slo-mo read, not normal speed..... but more on the normal mode later! :-)

The zone plates should show equal H & V resolution if it's read like the C300 .......

Yes indeed - but Adams chart for 240fps does show equal H & V resolution, and that's what I was referring to there. At 240fps I'm suggesting like the C300 - but only reading alternate blocks horizontally and vertically.

......but there is something odd about the processing or sensor readout in the FS cameras that results in asymmetric resolution at HD in normal modes.

Indeed. This has puzzled me as well for quite a while. If we accept that the SENSOR is symmetrical about the axes, with square photosites (which is reasonable, given what we know about future 4k plans), then it must mean it's something asymmetrical about the readout and/or processing.

I've literally just had a conversation which has come up with a plausible explanation as to what's happening - yes, it's speculative, but it's the only explanation I've heard which remotely answers observed facts. It's left me thinking "so obvious - why didn't I think of that!?!" :-)

Given a quad-HD sensor, it would seem most logical to read it in a manner similar to the C300. Why deBayer when a simpler process will give full 1080 4:4:4 results? (As proved by the C300.)

The suggestion as to WHAT is happening is that the FS700 does indeed read out in a similar way to the C300 - but uses 3x2 blocks to form each pixel rather than 2x2, three photosites horizontally, two vertically. (A few variations in detail were suggested, but fundamentally they ae all based on a 3x2 fundamental unit.)

That would imply you could expect vertical resolution up to about 1080 lpph (which seems to be what happens) and 720 lpph for horizontal resolution before aliasing. That also ties in with observed figures.

That's speculation, yes, but the only suggestion I've heard that seems logical. My immediate reaction was WHY? Yes, it seems to satisfy observed results, but why do it like that? What not just do what the C300 does and use 2x2 blocks?

At this point, your guess becomes as good as mine. Maybe simply because it's too difficult to process all the data at frame rates? Maybe just to deliberately downgrade the output to avoid a much cheaper camera rivalling the F3?

[Alister Chapman]

3x2 blocks with a bayer type CFA is possible but I can't see it as you would have different R and B sampling structure on alternate blocks which would be processing nightmare:

GRG RGR GRG RGR
BGB GBG BGB GBG

Doesn't make any sense to me. Possible yes, reasonable, no. For the sensor to work at 4K it almost certainly needs to have some kind of conventional CFA arrangement.

Looking at the noise figures, sensitivity and latitude differences between the FS100 and FS700 I think it's pretty safe to say the sensors are different. Sony have already stated the pixel counts as 3.43 MP and 8.3 MP respectively. The FS100 is clearly bayer, the FS700 maybe not, yet both cameras exhibit remarkably similar H resolution. I still think this is a processing quirk.

[David Heath]

Originally Posted by Alister Chapman

3x2 blocks with a bayer type CFA is possible but I can't see it as you would have different R and B sampling structure on alternate blocks which would be processing nightmare:

GRG RGR GRG RGR
BGB GBG BGB GBG

I disagree, the processing would be pretty easy, especially if pixel binning was used. (Where the charges from more than one photosite get combined together.)

In the two cases, each has 3 green photosites, and either 2 red and 1 blue, or 1 red and 2 blue. Consequently, it’s a case for each block of binning charges for each colour, then scaling by 1,2 or 3 to derive direct R,G,B values for each block.

In the previous post I did mention that “ A few variations in detail were suggested….” Two possibilities below – read photosites in bold, others skipped.

GRG RGR GRG RGR
BGB GBG BGB GBG

GRG RGR GRG RGR
BGB GBG BGB GBG

In these cases, the resolution is still based on 3x2 blocks, but in each case the basic read corresponds to a basic Bayer block – 2 green and 1 each red, blue photosites

For the sensor to work at 4K it almost certainly needs to have some kind of conventional CFA arrangement.

Well yes, but for 4k operation we are only talking about reading out basic RAW data, a value for each photosite. For this case I’d expect conventional deBayering to be done in post to maximize the achieved resolution, in which case it doesn’t need to be done in real time and all the processing power of a computer is available.

What I’m referring to above is use in 1080 mode. What’s the point in building in powerful processing to do that deBayer, when a far simpler method (direct read) can achieve 1080 far more simply?

Looking at the noise figures, sensitivity and latitude differences between the FS100 and FS700 I think it's pretty safe to say the sensors are different.

Hah, I was deliberately only talking about the FS700….. which is known to be 4k capable, and you yourself have said has a quad-HD sensor! But since you brought up the subject of the FS100……. :-)

Sorry, I’m becoming even more convinced they share the same (quad-HD) sensor. That said, then it’s highly likely that the exact form of processing differs slightly between them – hence the differences, in spite of the fact that certain key attributes are identical. As one suggestion, maybe one makes full use of the 3x2 blocks (reading all 6 photosites), maybe the other reads 2x2 blocks and then skips columns? Such as that would certainly mean a differing in sensitivity and latitude - whilst keeping broadly the same resolution characteristics.

Something like that may also explain artifacting Adam Wilt noted in his review of the FS100 - note particularly the "this only seems to be an issue on vertical or nearly-vertical edges" - the effect of column-skipping, maybe?:

Normally, the false-color artifacts are about a pixel or two wide or high (the minimum size visible using 4:2:0 recording), and fairly unnoticeable. However, a slow pan or tilt can cause affected details to twinkle or flicker between two colors (bouncing between a pale yellow-green and a pale magenta, for example); the flickering may call attention to what would otherwise be an insignificant artifact. Red/cyan false-color artifacts may also appear, though this only seems to be an issue on vertical or nearly-vertical edges.

Originally Posted by Alister Chapman

Sony have already stated the pixel counts as 3.43 MP and 8.3 MP respectively. The FS100 is clearly bayer, the FS700 maybe not, yet both cameras exhibit remarkably similar H resolution. I still think this is a processing quirk.

It’s my recollection that when the F3 came out a statement said that it was 3.4 megapixel Bayer, yes. (I’m not in any doubt about that.) But when the FS100 came out, the references went away for quite a while. I seem to remember asking someone at the time directly “do the F3 and FS100 share the same sensor?” The answer came back something like “they certainly both have an s35 sensor, yes”.

And from that, yes, I too then assumed it was logical they shared the same sensor. Regarding the FS700 (and in my view, the 100 as well) then if not Bayer, what? AFAIK, Sony have three basic architectures – Bayer, ClearVid, and Q67. It’s not going to be ClearVid – that’s optimized for small consumer sensors, squeezing the last ounce of sensitivity at the expense of colour resolution. And Q67s fundamental point is that there is one unique green photosite to every output pixel, so twice as many photosites in total as output pixels. If it is intended for 4k use, and is Q67, it follows it would have to be 2x3840x2160= about 16.5 megapixel. It’s not, it’s about half that and hence I’d say is pretty certain to be Bayer.

[Alister Chapman]

Still don't buy 3X2 readout. The colour aliasing would also be highly asymmetric skipping pixels like and the zone plates would show that up very clearly. That and it just doesn't make any sense to deliberately end up with asymmetric resolution and lower sensitivity just by skipping pixels. Clearly the FS700 sensor can be read fast enough to not need to pixel skip at normal speeds, otherwise it wouldn't be able to work at 4k.

The published pixel counts for the FS100 and FS700 are in any of the technical references for the cameras.
FS100 IS 3.3MP and FS700 IS 8.3MP in HD (which is quad HD). Sony have also very clearly stated that the sensor used in the FS700 is a new sensor. Sony have also publicly stated that the F3 and FS100 use the same sensor and the published specs are the same for both. I have no reason to believe Sony are lying or have any reason to try to mislead us. I'm more inclined to believe Sony's published figures than conjecture based on an anomaly that can't be explained by zone plates alone.

Sony : NEX-FS100E (NEXFS100E) : Technical Specifications : United Kingdom

There is no reason why clearvid or a variation of clearvid can't be used on a large sensor. It would bring exactly the same benefits and colour resolution is little different to traditional bayer. It could simply bayer rotated through 45 degrees. It could be Q67 if pixels refers to binned photo sites (although I doubt it). The one thing I have never seen published by Sony is that the sensors are Bayer.

Must invest in a decent zone plate chart.

[David Heath]

Originally Posted by Alister Chapman

Still don't buy 3X2 readout. The colour aliasing would also be highly asymmetric skipping pixels like and the zone plates would show that up very clearly.

Any colour aliasing will only be visible at high spatial frequencies, of the order of the photosite spacing - hence over 2000lpph. Any zone plates I've seen don't really go up that far. For frequencies of even 1000lpph the lines will spread over several photosites - so tend to even out as white/grey/black.

That and it just doesn't make any sense to deliberately end up with asymmetric resolution and lower sensitivity just by skipping pixels.

I look at that point the opposite way - it doesn't seem to make sense for the resolution to be asymmetric, period! But it is - that is what's measured - so why? 3x2 readout of quad-HD (or something based on 3x2) is the only explanation I've heard that remotely makes sense. At least, it complies with measurements, even if not answering why they chose to do it.

Clearly the FS700 sensor can be read fast enough to not need to pixel skip at normal speeds, otherwise it wouldn't be able to work at 4k.

Yes, at least for reading and straightforward RAW output. But what if the bottleneck is elsewhere, in the chip processing maybe? Conceivably, it's designed first and foremost for 4k RAW, the 1080 abilities are secondary. And even with direct-read it may be difficult to fully process at 60p.

The published pixel counts for the FS100 and FS700 are in any of the technical references for the cameras.

Odd things happened about 18 months ago. My memories are that initially figures were released for the F3, then withdrawn with Sony no longer commenting publicly at all for quite a long time. It's what I remember personally, and I've just found a reference from the time to back this up: http://www.dvinfo.net/forum/sony-xdc...ml#post1615837

Originally Posted by Carlos Molina

Interestingly, the two lines about sensor 'effective' resolution, and camera measured resolution have been deleted from the last version of the brochure.

These two lines don't appear any more:

Effective Sensor Resolution (HxV): Approx. 3,36 Mega Pixels

Horizontal Resolution (1920x1080/59.94i) 1000 TV lines or more

link to the old version of the brochure:
http://www.cameradepartment.tv/files...ads/PMW-F3.pdf

link to the more recent version of the brochure:
http://www.matrixvideo.ca/pdf_folder...3-BROCHURE.pdf

My recollection is that the public withdrawal of F3 specs coincided roughly with the time the FS100 started to be talked about.......

I have no reason to believe Sony are lying or have any reason to try to mislead us.

"Lying" implies both giving inaccurate information, and giving it deliberately. It's quite possible to give inaccurate information by mistake. As example, I was recently at my local garage and whilst waiting for paperwork the conversation with the salesman turned to future technology. He told me that his companies plans for next-gen cars were for them to be powered solely by water - he'd been to a presentation about it. When I expressed doubt, he took me over to a wall display to prove it - it turned out that he'd confused powered by HYDROGEN and producing water as a by product, with powered by water. Salespeople aren't always very good on technical matters.

There is also an obvious reason why Sony may have tried to be reticent about details of specs round this time - at least if you believe what I say about the FS100/F3 sensors differing.

*IF* the FS100 does have a quad-HD sensor, and *IF* they had volunteered full information about it early on , it would obviously have led to conjecture about a future 4k capable camera - exactly as happened after the C300 was announced. But the C500 wasn't far away at C300 launch, the FS700 was still a long way off at FS100 launch (and 4k ability even further away).

If you accept that, is it at all surprising that Sony would be coy about wanting to discuss FS100 sensor details? And since "does the FS100 have the F3 sensor?" then becomes the obvious question, then withdrawing F3 information makes sense. Better to say nothing than half a story. Don't forget they come from dfferent factories. It's easy to see a scenario where the F3 is launched with relevant details, the other factory sees the problems ahead and then asks for the F3 details to be publicised as little as possible.

There is no reason why clearvid or a variation of clearvid can't be used on a large sensor. It would bring exactly the same benefits and colour resolution is little different to traditional bayer.

In a block of 16 photosites, ClearVid has 12 green and two each red and blue (for Bayer the figures are 8,4,4) - it follows that the sensitivity is a bit better overall, but the colour resolution suffers considerably. For very small sensor, for consumer cameras, the compromise make sense. For large sensors for professional use it doesn't - the big sensors are easily sensitive enough anyway. ClearVid doesn't give colour resolution even up to 4:2:0

It could be Q67 if pixels refers to binned photo sites (although I doubt it).

The whole point about Q67 is that the number of green photosites equals the number of desired output pixels, and there are as many again red/blue in total. If the FS700 has about 8 megapixels, it's not Q67. It would have to have 16 megapixels, that's inherent to Q67.

Must invest in a decent zone plate chart.

What's becoming clear to me is that as cameras get better, to make sense of what's going on it's what's happening out of band that is becoming of interest. It used to be fine for a chart to register up to 1200lpph - it's becoming far more important to see what's happning in the 1000-2000 lpph range.

Of course, you can just half fill the frame with a 1200 lpp chart, I suppose...... :-)

[Alister Chapman]

Don't know why Sony published the sensor pixel counts, then withdrew them and now have them published again, but they did publish them, they are published again and they do tell us that the F3 and FS100 have the same 3.36 MP counts and the FS700 is 8.3 MP. I doubt the published figures are a mistake or error, they have been around for long enough for someone to have noticed by now. I suspect the FS100 sensor details were withdrawn simply to protect F3 sales. Why buy an F3 when you can get a camera with the same sensor for less money, I don't think it's any more sinister than that. The fact is the figures were/are published and I don't believe anyone is trying to mislead us either deliberately or otherwise. The one thing being overlooked is the massive amount of unused pixels on the FS700, an 11.6 MP sensor seems excessive and far more than the normal extra pixels used for black level and noise sampling. What else was this sensor designed for, is it a very big sensor for something else with just a central window used for video or does the camera use the full hight and width of the sensor and then pixel skip down to 8.3 MP?

[David Heath]

Originally Posted by Alister Chapman

I suspect the FS100 sensor details were withdrawn simply to protect F3 sales. Why buy an F3 when you can get a camera with the same sensor for less money, I don't think it's any more sinister than that.

To clarify, it was the F3 figures that were withdrawn – not the FS100. My memory is that no official statement was made about the FS100 at all around this time. I actually asked “will it be the same sensor as the F3?” and didn’t get a definitive reply. Because it was also s35, it was generally ASSUMED to share the s35 sensor of the F3. Since then, there was a lot of evidence to backup the initial reports about the F3 (3.4 megapixel) so the assumption that the FS100 was 3.4 mp rolled initially from that.

I certainly don’t think there was anything “sinister” intended, nor do I believe there was ever any intention whatsoever to deceive. Not saying anything is not “deceiving” – you can’t be described as misleading if you say nothing. The unfortunate matter may have been the F3 people initially giving specs of their camera (they had no reason not to) – by the time the FS100 team realized what was likely to happen, it became difficult to withdraw what had already been said. Subsequently, it’s not beyond the realms of possibility that more junior people at Sony were misled themselves by the popular assumption!

Originally Posted by Alister Chapman

The one thing being overlooked is the massive amount of unused pixels on the FS700, an 11.6 MP sensor seems excessive and far more than the normal extra pixels used for black level and noise sampling. What else was this sensor designed for, is it a very big sensor for something else with just a central window used for video or does the camera use the full height and width of the sensor and then pixel skip down to 8.3 MP?

Yes, good point. I had been thinking it was for a stills mode, that they were doing something similar to the GH2 in terms of area and aspect ratio – making most use of lens coverage area. Hence, for 4:3 stills use, the 4:3 coverage is more vertically than 16:9, less horizontally – it’s not just a 4:3 crop of the 16:9 area.

But looking at numbers I don’t think that works out. The brochure also states:

Total pixels approx. 11.6M
Effective pixels in movie shooting (16:9) approx. 8.3M
Effective pixels in still picture shooting (16:9) approx. 8.4M (3:2) approx. 7.1M

There is one other possibility. Whatever the truth about the FS100, I’m pretty sure the FS700 is 3840x2160 active photosites in 16:9 mode. That agrees both with the 8.3Mp quoted above, and what I previously said about the way the 240fps mode is read. (Reading out a 2x2 block out of a 4x4 block.)

But what if the total area of the chip had a 2.35:1 aspect area? (The FS700 just cropping the central 16:9.) If we assume 2160 vertically, that would make 5076 horizontally, and a total of 10,964,160 photosites altogether for active area? Add on the extra for black level and noise sampling, and that matches 11.6M pretty closely.

Alister, on a completely different subject, are you thinking of leading any excursions to try to see the Northern lights this coming winter - I know you did last winter?

[Alister Chapman]

However you do the math for the 11.6 MP of the FS700 it doesn't really appear to make a great deal of sense. 2.35:1 is an interesting thought, but what lenses would you use? You couldn't use PL glass it would be too wide, so you would have to use full frame DSLR lenses. Who would really want a 2.35:1 video camera that can't use PL glass?

I'm a northern lights expedition trip late December and the 2 in Feb 2013: Northern lights Expeditions to Norway. | XDCAM-USER.COM

[David Heath]

Originally Posted by Alister Chapman

2.35:1 is an interesting thought, but what lenses would you use? You couldn't use PL glass it would be too wide, so you would have to use full frame DSLR lenses. Who would really want a 2.35:1 video camera that can't use PL glass?

Fully agreed - the lens aspect had occurred to me as well. As for not many wanting it, I'm not so sure. If true it would give Sony the chance to bring out an "FS800" without too much extra development cost which could do all the FS700 does - and an extra big USP as well. That would include regular 4k (quad-HD and 4096x2160) with PL glass - and 5076x2160 with full frame glass.

Maybe we'll see 4k 2.35:1 monitors aimed at the consumer before too long? And that will drive the market for cameras to produce content.....? It's pure speculation, yes, but I can't think of a better explanation for the "why 11.6 MP?" question you pose. 2.35:1 would tie in very well with the numbers.

I'm a northern lights expedition trip late December and the 2 in Feb 2013: Northern lights Expeditions to Norway. | XDCAM-USER.COM

Thanks! The dates may be a little early, though the point about moon phases I was already aware of. I need to check a few things!

[David Heath]

Originally Posted by David Heath

Thanks! The dates may be a little early, though the point about moon phases I was already aware of. I need to check a few things!

Alister - you should have a reply from me via your website contact form.