hd 4:2:2 : lowest priced cam

[julien stiegler]

Ok,

 

i know this is the usual topic here but no post really gives an answer.

 

I just finished a music video shot in HDV i did the compositing in shake (adding maya backgrounds and animations to the bluescreened actors).

Obviously the keying on 4:2:0 was a nightmare and the result is not beautifull...

And obviously it was made for an SD broadcasting (the director didn't ask me before choosing the camera)

 

Now if tomorow the director want to shot another movie, what should i recommend for is low budget?

- Forget HD and use digital beta?

- use DVC pro HD (is there a cheaper alternative to varicam??)

- use film (but there is no scan facilities in 16 mm and 35 is very expensive)

 

Thank you in advance!

[David Mullen ASC]

Yes, you pretty much understand your alternatives -- avoid 4:1:1 systems like DV and HDV for chromakey work.

 

I'd also avoid interlaced-scan only cameras for the lack of a film look.

 

That sort of leaves the Panasonic SDX900 in DVCPRO50 mode, or one of the 24P Sony IMX cameras, for standard def, or the Varicam or F900 in HD. There will be the Panasonic HVX200 coming out by the end of the year, which is a consumer HD camera that records DVCPRO-HD.

 

And there is 16mm as well.

[drew_town]

JVC's DigitalS (D9) cameras and decks are buillt around 4:2:2 technology, and the gear is reasonably priced. List price is only $8000 for this camrea. $15000 for this model. JVC has a 16:9 native model as well. My coworkers shoot on the 16:9 model often.

[David Mullen ASC]

JVC's DigitalS (D9) cameras and decks are buillt around 4:2:2 technology, and the gear is reasonably priced. List price is only $8000 for this camrea. $15000 for this model. JVC has a 16:9 native model as well. My coworkers shoot on the 16:9 model often.

<{POST_SNAPBACK}>

 

Those are interlaced-scan only though, right? I'm not sure that standard def interlaced-scan is a good look for music videos and narrative movies...

[julien stiegler]

thank you for the replies !

 

 

So it seems like a varicam is still the cheapest way if we have chroma keys + hd mastering ... It costs $66 000 but if the number of keyed shoots is not eccessive we can rent the cam with the blue screen set and use another (cheaper and affordable) cam for pure live action shots without keying....

I think it's cleaver to rent good hardware for specialized works...

Am i wrong, is it how you work usually?

[David Mullen ASC]

Depends on what you are intercutting it with. If you are shooting interlaced-scan on a Sony PD170, for example, it would make more sense to shoot interlaced-scan on a Digital Betacam, for example, for chromakey work. If you are shooting 24P on something like a DVX100, then it would make more sense to shoot 24P on the DVCPRO50 SDX900 or the 24P HD on the Varicam and downconvert (but sticking to 4:2:2.)

[drew_town]

Those are interlaced-scan only though, right? I'm not sure that standard def interlaced-scan is a good look for music videos and narrative movies...

<{POST_SNAPBACK}>

That's right. HD would obviously be the better choice. However it's still a better choice than DV (progressive or not). I think it's easier to fudge the progressive look in post than match the clarity of a more professional format.

[Gary Robinson]

At NAB this year, I watched a demo of keying software called Ultra by Serious Magic (seriousmagic.com). Using this software, they pulled amazing keys even using a one chip dv camera. The latest version is hdv compatible. You might want to look into it. I picked it up, and it does a great job for only $400-$600 dollar software.

[julien stiegler]

Hey David,

 

 

i didn't notice before but the HVX200 sounds like a revolution!! HD 4:2:2 for less than 6000$ !

 

hey Gary!

i am sorry but i really don't trust 4:2:0 correctors... we can do similar things in shake you know (interpolate U and V using Y info) it can work sometimes but it's so dangerous if not in close up shots... I'll try ultra keyer anyway just to check, thank you.

[Lars.Erik]

Hey David,

i didn't notice before but the HVX200 sounds like a revolution!! HD 4:2:2 for less than 6000$ !

 

hey Gary!

i am sorry but i really don't trust 4:2:0 correctors... we can do similar things in shake you know (interpolate U and V using Y info) it can work sometimes but it's so dangerous if not in close up shots... I'll try ultra keyer anyway just to check, thank you.

<{POST_SNAPBACK}>

 

 

Yes, it will be interesting to see what the HVX200 will do. Progressive with DVCPRO-HD. But the problem is the price range, the limited amount of recording time and the lack of external lens.

 

This camera use the new P2 technology which records HD onto a small chip. The problem is that one chip costs over $1.000 and can only record 8 minutes. I'd rather go with 16mm than that. At least then you'd have 11 minutes of footage.

 

But then again one can always buy a harddisk and dump the footage on the chip onto it, erase the disk and then reuse it.

 

I don't know. It just sounds like a difficult way to have to work. But the DVX100a is a good camera. So who knows, I might just have to eat my own words here...

[Phil Rhodes]

Hi,

 

> The problem is that one chip costs over $1.000 and can only record 8 minutes.

 

Even less, with the smaller card recording HD.

 

Phil

[David Mullen ASC]

I was told that the largest P2 cards (8 gig?) holds 20 minutes each at 24P/720, and there are two card slots that switchover automatically, so that's 40 minutes.

 

Remember, these are like compact flash cards for a still camera, so you're meant to fill them up and empty them. So even though two or three cards is a big investment, if it allows you to shoot 24P/720 on DVCPROHD for under $10,000, that's still a lot cheaper than a Varicam and better quality than an HDV camera. You just have to think of it as an $8000 camera instead of a $6000 camera.

[Lars.Erik]

Hi David,

 

when you say dump the cards, will Panasonic have a "hard disk" that one can dump the footage on or will one have to have a computer available?

 

It does sound like a difficult way to have to shoot, or do you totally disagree with me?

[julien stiegler]

The Panasonic offers miniDV tape for recording, and P2 memory cards for high-def (and standard-def DVCPRO25 and DVCPRO50) recording. It can also transfer files from the P2 card directly to an off-the-shelf USB 2.0 external hard disk, or to a potentially-forthcoming FireStore type of device

Firestore ? sounds interesting

[John Allardice]

I was told that the largest P2 cards (8 gig?) holds 20 minutes each at 24P/720, and there are two card slots that switchover automatically, so that's 40 minutes.

 

Remember, these are like compact flash cards for a still camera, so you're meant to fill them up and empty them. So even though two or three cards is a big investment, if it allows you to shoot 24P/720 on DVCPROHD for under $10,000, that's still a lot cheaper than a Varicam and better quality than an HDV camera.  You just have to think of it as an $8000 camera instead of a $6000 camera.

<{POST_SNAPBACK}>

 

 

Panasonic are also releasing the camera at a $9999 price bundled with two 8Gb p2 cards, this'll give you 16 minutes of storage at 1080 24p, which this camera is also capable of. Once you've shot the cards can download to laptop through the PCMCIA slot( which the P2 cards plug right into) whilst you're preparing the next setup, and viola, you got the equivalent of a permanently self-loading 1500' mag

 

John Allardice

Digital DP

Digital Domain

[David Mullen ASC]

It does sound like a difficult way to have to shoot, or do you totally disagree with me?

<{POST_SNAPBACK}>

 

It depends on if I'm the guy who has to spend the time transferring the footage from the cards to a hard drive...

 

Shooting-wise (i.e. using the camera), I don't see any difference compared to tape -- it's the data transfer time that has to be factored into the shoot. A different way of working.

 

Obviously as the cards get cheaper or bigger, it gets more convenient, if you can save any file transferring until the end of the day, after shooting is over.

[Bob Hayes]

Can someone please explain to me what 4:2:2 and 4:1:1 etc. is. I used to know but I've been shooting film for the past couple of years.

[Peter J DeCrescenzo]

Can someone please explain to me what 4:2:2 and 4:1:1 etc. is.  I used to know but I've been shooting film for the past couple of years.

<{POST_SNAPBACK}>

 

I'm sure Phil R. & others may want to jump in here, but very briefly, I believe 4:1:1 video contains half the color information of 4:2:2 video.

 

Video formats have varying colorspace record/play capability. For example, NTSC DV video uses a 4:1:1 colorspace, PAL DV uses 4:2:0, DVCPRO-50 uses 4:2:2, and DVD-Video discs use 4:2:0. I believe Sony's new SR2 decks can record 4:4:4 color. But small colorspace numbers don't necessary mean ugly video: HDCAM uses a 3:1:1 colorspace.

 

There much more info on this topic on Adam Wilt's website; for example, about half-way down this page:

http://www.adamwilt.com/DV-FAQ-tech.html

 

... and on Michael Brennan's site; for example:

http://www.hd24.com/compression_numbers.htm

 

All the best,

 

- Peter DeCrescenzo

[Phil Rhodes]

Hi,

 

It's a ratio between the number of luminance (Y) plane samples and the number of colour difference (U/V) samples. 4:2:2 video has, therefore, two colour samples for every luminance sample. Standard def PAL, for example, might have a Y plane of 720x576 pixels, and UV planes of 360x576 pixels. NTSC DV has horribly low colour resolution horizontally of 180 pixels, with only one colour sample for every four luminance samples.

 

The strange 4:2:0 notation refers to subsampling in both the horizontal and vertical axes. Note so far we've been talking about samples on a line, intended to approximate the lower chroma bandwidth of non-digital formats like BetaSP. Digital video systems don't see them as scanlines, but simply as rows in a bitmap, and therefore space can be saved by subsampling both along lines and between them. A 4:2:0 picture therefore has half-resolution U data on one line and none at all for the V plane; the following line has half-resolution V data and no U. The device reconstructing the frame simply holds over the data from the previous line to create full UV planes. This is much closer to the way PAL (and even more so, SECAM) work and means that PAL DV has similar chroma resolution to Digital Betacam horizontally, and only half as good vertically. The received wisdom that you can't key DV mainly applies to NTSC for this reason.

 

Phil

[Peter J DeCrescenzo]

Hi Phil: Thank you for the additional information. I have a related question: Is using a description like "8-bit" or "10-bit" together with labels like "4:2:2" or "4:4:4", etc. redundant, or is bitdepth different from colorspace?

 

Apologies for what may be a dumb question.

 

- Peter

[Brian Wells]

Can someone please explain to me what 4:2:2 and 4:1:1 etc. is.  I used to know but I've been shooting film for the past couple of years.

<{POST_SNAPBACK}>

In English...

 

In 4:1:1 NTSC, the 4 refers to 720x480 green (aka luma, black/white) resolution.

The 1 and 1 refer to 180x480 resolution for red and blue channels.

 

In 4:2:2 NTSC, the 2's signify a 360x480 resolution in red and blue.

 

It's very simple to remember in those terms.

 

Here are some pictures of the visual differences between HD formats with various color sampling patterns and compression algorithms.

 

720P comparison:

http://www.icexpo.com/dvx100/720vs720.png

 

1080i comparison:

http://www.icexpo.com/dvx100/1080vs1080.PNG

 

Somewhere on the internet there must be examples of the SD variety... I don't remember where.

 

Hope this helps-

BW

[Phil Rhodes]

Hi,

 

> Is using a description like "8-bit" or "10-bit" together with labels like "4:2:2" or

> "4:4:4", etc. redundant, or is bitdepth different from colorspace?

 

Bit depth is different from what I think you mean by colourspace. The concept of recording the UV planes at a lower resolution than the Y plane is properly referred to as colour subsampling and affects spatial resolution, that is how well small details (or in this situation the chromacity of small details) are recorded.

 

The number of bits used to record the sample, that is the value (brightness) of one pixel within one channel, determines the accuracy with which that level of brightness is recorded. If, for example, the recording system uses eight bits, it can record the value as any one of 256 (2^8) levels. If the actual value read from the CCD is inbetween two of those levels, it is assumed to be equal to the nearest available level.

 

This characteristic is referred to as quantisation, and has two detrimental effects: first, it can cause banding in areas of finely graduated colour, particularly if postproduction colour correction is applied, and secondly if the value of a pixel fluctuates very slightly between frames, the pixel's value could be recorded as either of two adjacent quantisation levels over successive frames. This can cause unsightly flickering, and is called quantisation noise.

 

Both of these characteristics can be alleviated by recording the value of a pixel more finely, that is by using more bits. Every bit added doubles the number of quantisation levels, so 10-bit systems offer an accuracy of 1/1024 full scale. An important factor to understand, particularly when comparing DSP bitcout (often 12, 14 or even 16) with on-tape bit count (8 or 10, almost universally) is the concept of further quantisation due to colour correction. Whether that correction is applied as a postprodction grade or in-camera effect such as white balance, it has the effect of stretching the input sampling range such that some quantisation levels fall outside the output range and are lost; this can exacerbate quantisation problems as there are now effectively fewer levels, and those that do exist may be further quantised to whatever output range is in use. For this reason, DSP electronics use high bit counts to sample the (analog) CCD, ensuring that there are so many quantisation levels that even after a lot of mathematics, you still end up with reasonably accurate, lower bit count data on tape.

 

It's also worth understanding that bit count has nothing to do with dynamic range. It's possible to fit a film scan with 12 stops of range into an 8-bit image, it's just that those 12 stops will be comparatively coarsely sampled and colour correction, such as a DI, will suffer from quantisation problems. Dynamic range is controlled by characteristics of the CCD and the analog electronics that drive it.

 

Colourspace is, correctly, a term referring to the bounds applied to the technique used to describe a colour, generally in 3D space. RGB, YUV, LAB, HSV and others are common coordinate systems. I'm simplifying slightly, but in general RGB is a cartesian system and describes colour as a point within a cube; HSV is a polar system and describes colour as a point within a sphere. Because of these differences, cartesian and polar systems with identical absolute bounds occupy different volumes and can thus describe different areas of colour.

 

These issues are further complicated because each coordinate system has no inherent bounds - that is, the colour that results from a 0,1,0 RGB coordinate cannot be absolutely known without external data. It's going to be green, but what green? What wavelength, what CIE coordinate? Converting between systems - as a video camera does when it translates RGB sensor data into YUV data for the codec - is impossible without external data describing exactly what is meant by a given set of coordinates - effectively a standardised relationship between a notional coordinate system and real world colorimitery. As far as TV and film work go, the technical bodies have established what amount to standard bounds for RGB and YUV, among many others, allowing conversion between them and repeatable, standardised output - but the inevitable result of this is that there are areas of colour that are described by standard 601 RGB that 601 YUV cannot describe, and vice versa. Such colours are described as "out of gamut" and will be misrepresented.

 

So there you go. Subsampling, bit count, and colourspace. It's 20 past midnight and I have a shoot in the morning, so there's probably some drivel mixed in with the above.

 

Phil

[Peter J DeCrescenzo]

Thank you, Phil & Brian; very helpful.

 

So, if I understand this, when Sony says the SDI option card in my DSR-450WSL camcorder outputs "live" 10-bit 4:2:2 video, it means: The output signal results from 10-bits of quantisation and 4:2:2 colour subsampling. Also, the Y part of the colorspace is output at full resolution and the UV parts of the colorspace are output at half resolution (compared to 4:4:4). Is that correct?

 

(For what it's worth, Sony says the DSR-450WSL's ADSP "uses more than 30 bits in its nonlinear process, minimizing round-off errors".)

 

Concerning banding: Is a blue sky an example of where 10-bit video would be desirable vs. 8-bit video? Or gradulated reflections on a shiny surface, such as on an automobile?

 

Does banding tend to be more noticable in relatively bright or dark areas of a scene? For example, would 10-bit video produce smoother shadow areas compared to its affect on brighter areas? (I've noticed banded fades to/from black in various programs ... but I can't remember if I've mostly seen that on DVDs or broadcasts.)

 

Thanks again.

 

- Peter

Edited August 4, 2005 by Peter DeCrescenzo

[Phil Rhodes]

Hi,

 

This is where camera manufacturers' claims can get muddied with sales pitch, but what we hope has happened (and what has probably happened) is that the 10-bit SDI output is derived from the high-bit-count DSP data.

 

Blue skies can often have very subtle shading, so yes, that's where you'd look for trouble with quantisation. It's normally subtle even if present, and it's mediated a lot by random noise in the image, so it isn't something you'll see every day. To see it more clearly, get a DSLR image and start applying levels filters to it.

 

You will see more banding in fades to black on both DVD and digital broadcasting as there is no MPEG compression shortcut for "dimmer" - it just has to send new picture information, which stresses the codec.

 

Phil

[Michael Morlan]

Remember that, no matter what bit-depth the camera's processing package samples in, if you are recording to tape, your final image will be in that format's bit-depth. For instance, the JVC-GYDV500 has a 12/24-bit image processor. Too bad you lose all that when the data is recorded to the 8-bit DV25 tape.