Sony official brochure for the PMW F3 (just released)

I'm guessing that the sensor is oversampling the 1080p image with about a ~2440x1373 image since its a 3.36mp effective pixel resolution. Keep in mind that effective pixel resolution only measures the area of the sensor that is exposed to form the image, the actual sensor resolution is higher. For comparison the Arri Alexa's effective pixel resolution is 2880 x 1620. Larger photosites, but less resolution than an Alexa.

But for an S35 sensor we are talking about incredibly large photosites; huge full well capacity. This would explain the unreal clean images even with +18 db of gain. Obviously we haven't seen what dynamic range this camera can do as we haven't seen anything taken with S-log, but 12-stops seems to be not only reasonable but may actually be on the conservative side compared to how questionable DR measurements from different manufacturers are these days.
 
Agreed that having only 3.36 megapixels on a near-S35 sensor should produce massive pixels, leading to great DR and extremely clean low-noise images. And that's enough oversampling to produce a good solid 1080p image. But that would call into question whether it could deliver 4:4:4. How are they saying they'll get 4:4:4 out of a 3.6mpix Bayer sensor? For true 1080p 4:4:4 you'd need 1080p of red, 1080p of green, and 1080p of blue. That's a minimum of 6.6 megapixels.

If it's Bayer and 2440x1373 as Steve suggests, then that puts it as ~1220 x 690 in color resolution. Good enough for a solid 4:2:0, and maybe a bit more. How is Sony suggesting that's enough to make a 1080p 4:4:4 image?
 
Barry_Green said:
Agreed that having only 3.36 megapixels on a near-S35 sensor should produce massive pixels, leading to great DR and extremely clean low-noise images. And that's enough oversampling to produce a good solid 1080p image. But that would call into question whether it could deliver 4:4:4. How are they saying they'll get 4:4:4 out of a 3.6mpix Bayer sensor? For true 1080p 4:4:4 you'd need 1080p of red, 1080p of green, and 1080p of blue. That's a minimum of 6.6 megapixels.

If it's Bayer and 2440x1373 as Steve suggests, then that puts it as ~1220 x 690 in color resolution. Good enough for a solid 4:2:0, and maybe a bit more. How is Sony suggesting that's enough to make a 1080p 4:4:4 image?
Click to expand...

Barry, I posted a similar comment here:

http://www.dvxuser.com/V6/showthrea...on.-Enough-for-free-aliasing-1000TV-lines-444

But besides the capacity to capture uncompressed color, it is also written in the brochure that the camera can deliver an image of "more than 1000 TV lines" of horizontal resolution. I wonder if you can achieve that resolution if you are only sampling 1220 pixels in the horizontal dimension (green channel)?

Common sense will tell me that you need at least two pixels to capture 1 TV line pair (i.e. one cycle). Or to put it in other words, you need at least two pixels to capture two pixels. So to capture 1000 TV horizontal lines, you will need at least 2000 horizontal pixels. 1220 doesn't seem to be enough. Or is it that you can consider here all 2440 pixels (say, 1.3K green + 0.6K blue + 0.6K red) as contributing to the horizontal resolution TV lines output?

I thought, only the green channel was really responsible for image resolution in terms of TV lines since it is the channel that carries most of the information related to brightness. The other possibility is that the green interpolation works well enough that we can consider all horizontal 2440 pixels as contributing to the output image resolution, even though we are capturing 'real' green light (brightness) on only half of them.
 
Comes with: "PL lens mount adapter, Stereo MIC, Windscreen, IR remote, Shoulder Strap, CD-ROM (XDCAM Browser, SxS device driver software, PDF version operation manual), Operational Manual, Warranty card, PL lens kit (PMW-F3K only)"

NO Battery or charger??

Thats a bit sad. I really expected them to come with at least 1 battery and single charger.
 
Giuseppe Pugliese said:
Comes with: "PL lens mount adapter, Stereo MIC, Windscreen, IR remote, Shoulder Strap, CD-ROM (XDCAM Browser, SxS device driver software, PDF version operation manual), Operational Manual, Warranty card, PL lens kit (PMW-F3K only)"

NO Battery or charger??

Thats a bit sad. I really expected them to come with at least 1 battery and single charger.
Click to expand...

To be fair however, Sony are treating this camera like their high end models, which also dont come with batteries and charges. And due to the nature of the camera, many owners (especially rental houses) may not want the standard battery system anyway, opting for an AB Gold mount or V mount system.
 
lexicon said:
I wonder if you can achieve that resolution if you are only sampling 1220 pixels in the horizontal dimension (green channel)?
Click to expand...
Well, that's not entirely an accurate perspective though -- the luminance is made from all the colors. Using 2440 wide should be enough to provide a fully-resolved 1920x1080 HD image with minimal aliasing, that part I wouldn't even worry about. I'm just curious about the 4:4:4 coming from the single chip. I imagine it must be possible, I just don't know how.

Common sense will tell me that you need at least two pixels to capture 1 TV line pair (i.e. one cycle).
Click to expand...
Well, actually, you need four if you want it fully resolved. That's why the Nyquist frequency is what it is -- you'd have to sample double if you want to accurately capture the full essence. Two pixels could sample one pair of black and white lines, but only if the sensor was perfectly aligned. What happens if you pan half a pixel one way or the other? Uh-oh, now you have gray, with zero lines resolved! But, as said before, the color pixels can come into play here, the demosaic process doesn't just use red for red and green for green, the signal is created from a mixture of about 60% green, 30% red and 10% blue (ballpark). So it all comes into play.

Or is it that you can consider here all 2440 pixels (say, 1.3K green + 0.6K blue + 0.6K red) as contributing to the horizontal resolution TV lines output?
Click to expand...
Yes, but it gets more complex than that, because TV lines aren't typically measured like you're thinking. TV lines are measured as lines PER PICTURE HEIGHT. So resolution isn't typically measured on the full frame, it's measured in terms of a square from the center of the frame.

I've seen footage from the F3, I have no doubt it's resolving a very healthy 1080p. I just don't know how they can get 4:4:4 out of it. If it had three chips of 3.36mpix each, then yeah, no problem, that'd be perfect. But one single chip, with only 3.36mpix total, doesn't immediately explain how they get it. The Sony F35 has a 12.4 million pixel imager, which is pretty ideal for getting 1080p 4:4:4 -- it has an "RGB Stripe" pattern on it, where there are 1920 red, 1920 green, and 1920 blue pixels on the chip, in vertical stripes (i.e., no "pattern" like in a Bayer RGBG system). With 12.4 million pixels, they have an individual sensor of red, green, and blue in every possible position (well, actually, they have twice as many as strictly necessary because the pixels are square, so it's 1920x2160 red, 1920x2160 green, and 1920x2160 blue, so they supersample on the vertical).

Anyway, point being, that's how their other S35-sensor camera does it, same as the Genesis, and it delivers a true 1920x1080x4:4:4, so I don't know how the F3 can do the same job with 1/4 as many photosites. I could see it happening with 6.2 million pixels (because the F35 does supersample on the vertical, it has twice as many pixels as strictly necessary) but even then, that's still twice as many as the F3, so ... what gives?
 
I'm actually surprised to see a mic included. But we should keep in mind this is the Canadian brochure. Worldwide offerings may vary. I recall reading somewhere that in India it will be sold with a SxS card, battery, and charger.

I expect the US offering to match the Canadian one, although i'm still not sure whether we'll get the mic in the US...
 
Thanks Steve for your reply on the other threat. I will keep my comments here to keep everything in one place.

So, can we say that we have at least three distinctions here?

1. The total number of pixels (photosites is a better word) in the sensor (6.6 megapixels)
2. The total number of "effective pixels" (3.36 megapixels)
3. The number of pixels that the output digital image contains (HD: 2,1 megapixels)

F3 has a Bayer sensor as far as we know. I am wondering if what they mean by 'effective pixels' in this context refers to the total number of pixels that are capturing information (relevant to resolution) from the outside world, i.e. real information, not counting the pixels in the sensor that are inferring that kind of information though a process of interpolation from neighboring pixels. Assuming that the information that is relevant for resolution is mostly associated with the green channel (brightness), 'effective pixels' on a Bayer sensor will simply mean here the total number of 'green pixels'. Following this definition, and since on a Bayer sensor half of the pixels are dedicated to green, we will get a count of 3.3 effective pixels if our sensor has a total of 6.6 megapixels, a number that makes sense based on rumors and Sony comments that it is a 3.4K sensor.

The comment that the "effective pixel resolution only measures the area of the sensor that is exposed to form the image" sounds problematic to me, since I assume that all the area of the sensor is exposed to form the image (unless you are talking about some form of cropping factor which is not the case here). However, it still makes sense to distinguish between the part of the sensor that is exposed to 'real' information from the outside world (incoming light) versus the part of the sensor that is somehow 'inventing' that information through a process of interpolation. And maybe this what is behind the notion of 'effective pixel'. A term necessary to distinguish 'true' information from 'inferred' information. So, in the case of a Bayer sensor, 'effective' pixel resolution will simply represent the total number of pixels that are capturing true information that is relevant to resolution, in other words the total number of green pixels on the sensor.

With this definition of 'effective pixels' as 'the total number of green pixels of a Bayer', Barry's (and also mine) concerns would also be addressed because then, the total horizontal count in pixels will be in effect ~3.4K (6.6 megapixels) and not 2.4K (3.36 megapixels). I understand that as a rule of thumb, to capture uncompressed chroma HD (2K) on a Bayer pattern you have to double that value (4K) in terms of the total horizontal sampling resolution. Our estimated number of ~3.4K is relatively close to 4K, so it comes pretty close to 4:4:4. The other rule of thumb stays that in a Bayer pattern horizontal luma resolution amounts to 75% of the total horizontal resolution. If the total is 3.4K this gives us an horizontal luma resolution of 2.55 pixels, enough to capture "more than 1000 TV lines". But if the total horizontal resolution is only 2.4K then the luma resolution will be 1.8K which doesn't seem enough to produce images with "more than 1000 TV lines".

Anyway, I am assuming here that 'effective pixels' on a Bayer sensor -when it counts to measure resolution- basically means 'the green pixels on the sensor'. And this assumption could be perfectly wrong.
 
Last edited:
I get the feeling their idea of what 3.36 megapixels is, is different than what we are assuming. It's too far from being able to do 4:4:4:.
 
Barry_Green said:
Agreed that having only 3.36 megapixels on a near-S35 sensor should produce massive pixels, leading to great DR and extremely clean low-noise images. And that's enough oversampling to produce a good solid 1080p image. But that would call into question whether it could deliver 4:4:4. How are they saying they'll get 4:4:4 out of a 3.6mpix Bayer sensor? For true 1080p 4:4:4 you'd need 1080p of red, 1080p of green, and 1080p of blue. That's a minimum of 6.6 megapixels.

If it's Bayer and 2440x1373 as Steve suggests, then that puts it as ~1220 x 690 in color resolution. Good enough for a solid 4:2:0, and maybe a bit more. How is Sony suggesting that's enough to make a 1080p 4:4:4 image?
Click to expand...

"Effective" sensor resolution is always a bit troublesome (or just plain honest depending on the way you look at it). A 3 sensor (3ccd/3cmos) cameras are marketed as 1080p (1920x1080) even though it has thrice the 'pixels' (photosites) as a bayer. While it may have around 6 megapixels measured via photosites its 'effectively' 1080p.

While a rose by any other name would smell as sweet, a pixel by any other name does not: http://media.panavision.com/ScreeningRoom/Screening_Room/Demystifying_Part1.html

As it stands right now we don't know that this sensor is bayer CFA, there has been suggestions that it is not. Not enough is published right now, I've criticized Sony before with being too enigmatic with their specifications and this seems to be the case again here.

The sensor itself is obvious not a 3.36mp sensor, a larger 'surround view'/'look around' pixels are required for monitoring. Then there is the whole sensor itself, a lot of what's on the sensor is not actually used for image directly but are equally important, a lot of it is used for calibration purposes.

As an example the Arri Alexa has a 2880 x 1620 pixels effective pixel out. The surround view is 3168 x 1782, the actual Alexa sensor itself is actually 3392 x 2200. What we know about the F3 is that the 'effective' pixel area is 3.36mp, which I estimate to be 2440x1373 if its a bayer. Previously the sensor was said to be 3.43k; which it actually very well might be.

So when predicting 444 representation I personally can't say without knowing exactly that its bayer or not. But assuming that it is bayer CFA, for the sake of this conversation, we should expect it to be in a similar situation as the Alexa with a lower RGB resolution, in the sense that it won't be 'true' 444.
 
Viddovation said:
I'm actually surprised to see a mic included. But we should keep in mind this is the Canadian brochure...although i'm still not sure whether we'll get the mic in the US...
Click to expand...

I wouldn't get too excited about the mic. It is the standard mic that Sony issued with PD-150, Z1Us, etc. I suspect most everyone will be placing it in a spare parts drawer after they unbox the camera.
 
Hah, totally agree. I expect any mic they give me to suck.

I'm buying this sucker for the picture quality and the XLRs in. I've got better ENG mics than they could afford to include, and on narrative shoots that's all the audio guy's problem, not mine.
 
metalalien said:
I get the feeling their idea of what 3.36 megapixels is, is different than what we are assuming. It's too far from being able to do 4:4:4:.
Click to expand...

4:4:4 describes how sensor data is encoded to YCbCr for output, not how it is sampled and processed from the photosites before encoding. Color interpolation on a higher resolution over sampled Bayer pattern can produce full 1080p 4:4:4 encoded color resolution with less than 6 megapixels. It has to do with the diameter of the smallest focused dot, or COC, a 1080p output pixel can resolve for the sensor in question. If you can fit a full 4 pixel Bayer block of 2 G,1R, 1B inside that 1080p minimum COC pixel diameter, you have enough information to fully resolve 4:4:4 color.
 
Bayer doesn't require a 2x2 block for each pixel in order to resolve it, but if you did have that, it'd require an almost 9mp sensor to do 1080p @ 4:4:4.
 
Barry_Green said:
Agreed that having only 3.36 megapixels on a near-S35 sensor should produce massive pixels, leading to great DR and extremely clean low-noise images. And that's enough oversampling to produce a good solid 1080p image. But that would call into question whether it could deliver 4:4:4. How are they saying they'll get 4:4:4 out of a 3.6mpix Bayer sensor? For true 1080p 4:4:4 you'd need 1080p of red, 1080p of green, and 1080p of blue. That's a minimum of 6.6 megapixels.

If it's Bayer and 2440x1373 as Steve suggests, then that puts it as ~1220 x 690 in color resolution. Good enough for a solid 4:2:0, and maybe a bit more. How is Sony suggesting that's enough to make a 1080p 4:4:4 image?
Click to expand...

Thanks Barry for your comments and replies. I always learn a lot from them, as I do from the ones Steve Castle post too.

In relation with the F3 sensor-resolution conundrum I can comment this:

If you simply count which photosites are filtered, half of the Bayer sensor has green-filtered photosites, and one-quarter red and blue each. Assuming that the expression 'effective resolution' in the brochure refers, not to the total number of photosites on the sensor, but the number of actual pixels (i.e. units with full luminance and colour information), then we can imagine what will happen if we conceive 'effective resolution' as resolution after debayering.

As a general rule in the case of a Bayer sensor, you can consider that the final measurable resolution of the debayered image as roughly 75% of the original (for example, Adam Wilt found that the RED One measurable resolution was around 3,2K for a sensor of 4K. Graeme Nattress from RED obtained a similar value).

So, the F3 brochure says that the effective resolution of the sensor is 3,3 megapixels. Following Steve suggestion this will correspond to 2,4K horizontally (2440x1373). Applying our general rule of the 75%, we have that 2,4K / 0,75 gives 3,2K. So the original undebayered horizontal resolution could be around 3,2K, or 3200x1800, which is 5,76 megapixels.

5,75 megapixels seem to be a big enough oversampling number to get the desired uncompressed color sampling (4:4:4).

All of this makes sense as far as we don't consider the number of 3,3 megapixels (given in the brochure) as the total number of photosites on the sensor, but instead as the total number of pixels after debayering (the measurable resolution).

If the number 3,3 megapixels really refer to photosites (and not pixels), well there is still the possibility that it doesn't refer to the total number of green, blue, and red photosites (the totality of the sensor), but only the total number of green photosites. This, since information about luminance -which is relevant for definition- is mostly (not totally as you said) contained in the green channel. The green photosites on a Bayer sensors represent half the total photosite count, so a number of 3,3 mega will translate into 6 mega for the total sensor, and this seems enough to get uncompressed color 4:4:4.

Finally, there is still the possibility that the sensor really has a total number of 3,3 photosites (counting all, green , red, and blue). In this case I have no idea how you can get 4:4:4 from such a small number.
 
lexicon said:
All of this makes sense as far as we don't consider the number of 3,3 megapixels (given in the brochure) as the total number of photosites on the sensor, but instead as the total number of pixels after debayering (the measurable resolution).
Click to expand...
I would not be inclined to make that assumption at all though. This brochure is likely produced by the same marketing team that famously listed the EX1's LCD resolution as "1920x480" when it was, in fact, 640x480. They counted the red, green, and blue sub-pixels individually to come up with 1920x480.

If this was a white paper, you might be on to something, but it's a brochure from the marketing division, and as such I would be inclined to assume (yes, there's that word, "assume") that they are using the biggest numbers to make the best-possible marketing case. I think the definition of "effective" pixels being those that are used in the image-making area and excluding those used for look-around and black levels, is probably more accurate, although the word "effective" is quite odd there, in that they used to use the term "active pixels" to describe that.
 
A demosaiced image from a Bayer sensor has exactly the same number of RGB pixels as the number of active photosites on the sensor. People continue to confuse optical resolution measured in lines per picture height from a chart with the pixel format of the image file. A 4 pixel Bayer block, 2G-1B-1R, demosaics to 4 fully defined RGB pixels in whatever image format you convert it to. No pixels are lost unless your crop or scale to a lower frame resolution.

A 3.3 Mp Bayer sensor should optically resolve close to full 1080p without aliasing when debayered and scaled down to HD, probably at least 900 lines vertical, on par with an F800 3-chip camera more or less. Keep in mind the F3 is still a 1080p HD native format video camera as far as what it spits out the back end.
 
Last edited:
You must log in or register to reply here.
Back
Top