Sony F3 XDCAM 35MB/s and ProRes 422 220MB/s Comparison (Matthew Allard)

[lexicon]

Mathew Allard:
"A simple comparison between recording 4.2.0 35Mb/s as opposed to 4.2.2 10 bit Apple ProRes 422 (HQ) to a Cinedeck Extreme.

Amazingly the 35MB/s codec stands up much better than you think it would. You should never judge a codec purely by a number. 4.2.2 10 bit is sharper and slightly cleaner but it is very hard to pick the difference.

The night time shots were all shot on +12DB."


http://vimeo.com/21432202

[davidsoll]

Thanks for posting the side by side, very interesting. Did you have any time to grade the two and compare when each fell apart?

In addition to grading, I'm most interested in how well the XDCam codec holds up with effects. Obviously 10bit 422 will make rotoscoping and compositing much easier, but has anyone done any motion tracking with the internal codec? I'll be testing in the coming weeks, but if anyone has any experience with it already (ideally a direct comparison on how well a point tracker deals with XDCam vs off board format), it would be great to hear.

[tom.wong]

i've already done this test, and xdcam ex is very good at getting your image close to what you shoot in camera. WYSIWYG for the most part. but it really gets smoked when you have to start pushing the image in the grade. it won't even come close. you'd also be surprised on how big of a quality difference there is between 10 bit pro res, and 10 bit uncompressed when you really have to stress an image in post.

[Lenilenapi]

This is what I expected but it brings up a question about gamma.
If you are staying in EX 35mb are you better off shooting a WYSIWYG gamma like 709 or using cine gammas. Will the cinegammas weaken with the grading, or does this only apply if you are doing significant color rebalancing or trying to save something under or overexposed?

[Lee Saxon]

But how does 35Mbps compare in action/movement-heavy, non-locked-down shots that actually stress it?

[ddh]

In my experience with the EX3/NanoFlash recording to both simultaneously, out of camera footage looks exactly alike. When I've taken 35mb into apple color to correct IR (which I did on a recent shoot) the 35mb codec falls apart rapidly. If it's run and gun interviews with no deep correction needed ... the 35mb codec is superb. It's clear to me that the F3's clean image will mate beautifully with a 8bit or 10bit recorder at 4.2.2 ... it's the only way to go and is certainly necessary for broadcast work.

[Nate Weaver]

XDCAM, or even most compressed 8 bit formats work pretty well if the corrections you're making involve mostly mid tones and highlights. There's enough bits up there, so to speak, and there's no camera noise up there either.

But start messing with the color of the shadows, levels of the shadows (especially crushing them) or stretching the gamma far up or down to save exposure, and things get nasty really quick.

XDCAM works great for people that don't color correct, even these clips amaze me for what it CAN do, not what it can't.

[avro]

XDCAM, or even most compressed 8 bit formats work pretty well if the corrections you're making involve mostly mid tones and highlights. There's enough bits up there, so to speak, and there's no camera noise up there either.

But start messing with the color of the shadows, levels of the shadows (especially crushing them) or stretching the gamma far up or down to save exposure, and things get nasty really quick.

XDCAM works great for people that don't color correct, even these clips amaze me for what it CAN do, not what it can't.

Agree.
Might add make sure you don't shoot any Sport like Surfing.....
Thats why 50/60P recording in fs100 will be better

[davidsoll]

Cool. Beyond grading, has anyone messed around with comparing effects work like rotoscoping or motion tracking yet? Curious...

[Lenilenapi]

What do you consider the most important part of an upgraded recording. 4:2:2 or 10 bit. My guess is that 10bit 4:2:2 would be a big step up from 8bit 4:2:2 hence its better to get a mini Kipro or the upcoming SDI Atamos Samurai

[Gabrobot]

What do you consider the most important part of an upgraded recording. 4:2:2 or 10 bit. My guess is that 10bit 4:2:2 would be a big step up from 8bit 4:2:2 hence its better to get a mini Kipro or the upcoming SDI Atamos Samurai

Depends on what you're doing. For grading, it's definitely 10bit...chroma key work on the other hand would probably benefit more from 4:2:2. (Though I'm not sure you'd ever have 10bit, and not at least 4:2:2)

[Lenilenapi]

I'm asking because I'm thinking of what's an important addition . Nanoflash offers 4:2:2 but no 10 bit. Better to hold out for Samurai or Ki Pro methinks.
Actually they're both cheaper. I think Convergent should have gone for a 10 bit upgrade on the Nano instead of putting all their eggs in the more expensive and file heavy Gemini. Both would be best.

[Barry_Green]

I would have loved to see an AVC-based 10-bit Nano, and I still think there's room in the market for it. To be fair to them, the MPEG-2 hardware that they use only supports 8-bit, so they couldn't have made it 10-bit without designing an entirely new product based on new hardware, and ... there's still a ton of HDMI and HD-SDI cameras out there that are 8-bit only. All the Panasonic AG's (except the 300/370 of course), all the Canon cameras, all the JVC's, and all the HDMI cameras, are all 8-bit only. 10-bit is a rarity, only some of the Sonys, the HPX300/370, and the 2/3" cameras have 10-bit outputs, so ... I don't know that the Nano being 8-bit only is that bad, but I don't know that it would be my first choice for a true 10-bit head like the F3.

As for which is more important, 8-bit or 10-bit, keep in mind that almost all the HD codecs on the market in any camera, are all 8-bit, and always have been. The only 10-bit codec in a camera is AVC-Intra. Many/most pro codecs have usually been 4:2:2 (being DVCPRO-HD, AVC-Intra, DVCPRO, DigiBeta, XDCAM-MPEG422, etc) but we've been living with 8-bit almost entirely (HDCAM, DVCPRO-HD, XDCAM-everything, NXCAM, HDV, AVCCAM, DV, DVCPRO, DVCAM, DVCPRO50, etc). Doesn't necessarily mean much other than that it seems like a reasonable extrapolation would be that when the manufacturers have been designing codecs and systems, they have made room for 4:2:2 but rarely have they put in 10-bit. Also note that major broadcasters like the BBC and Discovery have approved 4:2:2 codecs for acquisition, editing, and mastering, but they don't specify 10-bit (i.e., 8-bit XDCAM-MPEG422 is acceptable, but 8-bit 4:2:0 XDCAM-EX is not; 8-bit 4:2:2 DVCPRO-HD is acceptable, but 8-bit 4:2:0 AVC/HDV are not. There are of course other reasons why HDV didn't make the grade and AVCCAM seems in the same boat, but the fact remains that they've accepted 8-bit in the past for unlimited acquisition, editing and mastering).

Also, when offered a choice, the manufacturers seem to have always chosen 4:2:2 over 10-bit, if they could only implement one. In other words, there are plenty of examples of 4:2:2 8-bit, but I can't think of any examples of 4:2:0 10-bit. Oh, wait, yes I can -- AVC-Intra 50mbps is 4:2:0 10-bit, but ... I'm not a fan of AVCI-50, I think it's way too compressed and doesn't look as good as DVCPRO-HD and I'd argue that even AVCCAM can frequently/usually look better than AVCI-50; I wonder if the codec wouldn't look better if they'd reallocated those bits and made it an 8-bit version and used the extra space for overall compression quality improvement. And, of course, it's also worth noting that the camera manufacturers seem to think that 8-bit 4:2:0 is adequate; all the XDCAM-HD and XDCAM-EX and NXCAM and AVCCAM products are all 8-bit 4:2:0. Doesn't mean that we have to find it acceptable, I'm just saying that -- the manufacturers seem to have made their choice (i.e., 4:2:2 is more important than 10-bit) but that doesn't mean you have to agree.

4:2:2 is readily available, but 10-bit is limited to AVC-Intra (and, when the memory recorder is released, HDCAM-SR). 10-bit is great in post, but hasn't really proved that big of a deal in widespread camera usage because you only really see the benefits of 8-bit if the noise level of the camera is extremely low. If the signal has noise in it it erases much of the acquisition benefit of 10-bit. Someone did a test on an EX1, recording a gradient through 10-bit external HD-SDI and found that it was indistinguishable from the 8-bit internal, because the noise in the signal covered up the gradation distinction that 10-bit offers.

For something like the F3, which is an extremely clean sensor, it would be entirely reasonable to expect that it should be able to take full advantage of all that 10 bit recording has to offer. 10-bit really shines in gradients and in post, when modifying contrast or brightness or pushing and pulling you will probably see the benefit of 10-bit clearly on steeper grades. I've said it before and will say it again, if I was getting an F3, I'd definitely get at least a Samurai, to take advantage of the capabilities that the camera has that XDCAM-EX doesn't (being, 10-bit 4:2:2). I don't think the Nano would be the best match for an F3, if going with a Convergent recorder I'd go with the Gemini instead. If going with a non-upgraded F3 (so no need for 4:4:4 etc) I'd take a real good, hard look at the Ki Pro Mini for the F3. The 10-bit 4:2:2 and HD-SDI are good matches, and it's affordable, and if you're running a bigger camera like the F3 you're more likely to be using 12v batteries which solves the powering issue for the Ki Pro Mini. Seems like a realy good combination. Gemini for the upgraded F3, Ki pro Mini (or, if it proves itself, Samurai) for the non-upgraded F3.

As for the initial post in this thread, I'm not surprised at all. On that type of shot, XDCAM-EX will perform quite well; we've been finding on the AF100 board that for many/most shots it's pretty difficult to tell the difference between NanoFlash-recorded high-bitrate footage, and the same footage recorded by the onboard AVCHD codec. And I found in my comparison between AVCHD and XDCAM-EX that as long as the shot doesn't stress the codec to its limits, it does perform quite well, so, yeah, this isn't surprising. If you want to demonstrate the differences, look at the tests that the Convergent Design folks did between the EX1 with and without the Nano, they ran some tests that show when the onboard codec reaches its limits and how the higher-bitrate Nano can retain more detail.

But yeah, you're never really gonna see that difference in a talking head type of shot; for that the XDCAM-EX codec is totally going to perform fine. (although, again, from a 10-bit 4:2:2 head you'd still get better/nicer footage from an external recorder that could take advantage of what the F3's head is capable of delivering).

[TimurCivan]

I'm asking because I'm thinking of what's an important addition . Nanoflash offers 4:2:2 but no 10 bit. Better to hold out for Samurai or Ki Pro methinks.
Actually they're both cheaper. I think Convergent should have gone for a 10 bit upgrade on the Nano instead of putting all their eggs in the more expensive and file heavy Gemini. Both would be best.

well, thats what the KIpromini and Samurai are for. No sense throwing ANOTHER hat in the ring when the market is small as is. CD made the right choice in going for the top end. Think of uncomressed 444 sLog as a kind of "raw". its capturing uncomressed color and 10bit depth, essentially everything the sensor is putting out ( sLog) and you then choose what to compress to later on.

[Lenilenapi]

Thanks, Great reply Barry. Kind of what I figured.
My guess is a lot of my clients won't want the greater post intensiveness of 4:4:4 but 10 bit 4:2:2 they might ask for. of course frankly I don't know how much difference those will make in post, so that is another question.

[prepost]

sorry to wake up dead (or at least sleeping) threads :)
I just posted this quick comparison test in my blog between footage from F3 and KiPro

its a short test, but it shows well where differences hide, and where native codec breaks down

also, its my first post in this forum, which i've known and appreciated for many years.

http://prepostconsulting.com/en/english-external-recorders-part-3-kipro-vs-sony-f3/ (http://prepostconsulting.com/en/blog/)


hector



hector berrebi
production and post consultant
www.prepostconsulting.com (http://www.prepostconsulting.com)

[kolak]

sorry to wake up dead (or at least sleeping) threads :)
I just posted this quick comparison test in my blog between footage from F3 and KiPro

its a short test, but it shows well where differences hide, and where native codec breaks down

also, its my first post in this forum, which i've known and appreciated for many years.

http://prepostconsulting.com/en/english-external-recorders-part-3-kipro-vs-sony-f3/ (http://prepostconsulting.com/en/blog/)


hector



hector berrebi
production and post consultant
www.prepostconsulting.com (http://www.prepostconsulting.com)


Yep- this show what 35Mbit MPEG2 realtime encoding is about.
Now take it to some eg. street etc (where whole frame is "busy") and you have idea how it's going to look like.
It's almost waste of F3 quality to shot on internal cards, but also in some case it's good enough and saves problems related to recorders.

[starcentral]

Thanks for posting the side by side, very interesting. Did you have any time to grade the two and compare when each fell apart?

In addition to grading, I'm most interested in how well the XDCam codec holds up with effects. Obviously 10bit 422 will make rotoscoping and compositing much easier, but has anyone done any motion tracking with the internal codec? I'll be testing in the coming weeks, but if anyone has any experience with it already (ideally a direct comparison on how well a point tracker deals with XDCam vs off board format), it would be great to hear.

I've motion tracked even crappy footage using built-in tools in AE, as well as planar tracking with Mocha (which shipped included with CS5) without any problems. I can't see the codec really playing into motion tracking at all, unless you are really tracking on the smallest of points in your footage but generally you would track on an object with some distinct edge and the results are really good. I've played around a lot with it using P2 and even miniDV footage without any issues. Hopefully soon I will play with F3 footage too.

[FilmDude]

Am I the only one that noticed that the motion jutter looks different between the 2?

[arturrito]

Don´t forget Pix240! another good option to get full F3 sensor capabilities

[Peter G. Johnson]

Honestly, Matthew should be presenting this sort of stuff to seminars and conferences. He explains things with clarity and a straightforward approach with examples that you can see and compare. That just about completes the picture.

[kolak]

As for the initial post in this thread, I'm not surprised at all. On that type of shot, XDCAM-EX will perform quite well; we've been finding on the AF100 board that for many/most shots it's pretty difficult to tell the difference between NanoFlash-recorded high-bitrate footage, and the same footage recorded by the onboard AVCHD codec. And I found in my comparison between AVCHD and XDCAM-EX that as long as the shot doesn't stress the codec to its limits, it does perform quite well, so, yeah, this isn't surprising. If you want to demonstrate the differences, look at the tests that the Convergent Design folks did between the EX1 with and without the Nano, they ran some tests that show when the onboard codec reaches its limits and how the higher-bitrate Nano can retain more detail.

But yeah, you're never really gonna see that difference in a talking head type of shot; for that the XDCAM-EX codec is totally going to perform fine. (although, again, from a 10-bit 4:2:2 head you'd still get better/nicer footage from an external recorder that could take advantage of what the F3's head is capable of delivering).

8bit is not a "big" problem, as MPEG2 compression is. You can live with it, but massive macroblocking is never nice on your main master.
In the same time 8bit is more visible with F3, because it's so clean. With many other cameras 8bit "issues" are covered with noise, but with F3 they are much more visible.
Same applies to Alexa, Epic- I do loads of feature films encoding and film based movies almost never cause a problem (due to grain). Digitally shot productions need way more attention and in most cases proper dithering during 10 to 8bit conversion is needed.
If you can, keep 10bit workflow, as it gives more headroom for grading and creates much more "reliable" final master. It's absolutely not true that 10bit master makes no difference when your target will be 8bit, eg Blu-ray disc. It does make massive difference- just depends how you use your 10bit main master, but this is separate issue. With a proper dithering 8bit version can look very good in terms of gradients where "fixing" banding on 8bit sources is not easy at all.

[klfi]

at the moment i am only able to record s-log on 8bit 4-2-0 sxs card. the desicion which external rec i will use in future is not done - but i will after NAB 2012.

I use Win7-PC, PPro and AE, Blackmagic and MB-Looks, not BM Resolve light becouse of less knowledge at the moment.

May i ask for my further understanding in grading and colour correction, does the following workflow have advantages:
1.record s-log 8bit 4-2-0
2. then transcode the footage to 10 bit 4-2-2 with AME (for examle to DNxHD422 or QT mov 10bit422, Prores is not possible becouse of PC-version)
3. then grade the transcoded exported 10 bit footage in PPro or AE
4. then export again to 8 bit 420 for example to use Blue-Ray or LCD-TV streaming clips

where would the advantages be ?
Alister Chapman once had a blog on his side about this workflow, but for me it was not final understandable. footage 8bit 420 - blow up to 10bit422 to grade or CC - and go back to 8bit 420.
Will (8bit-footage) macroblocks and noise not also be blown up in a bad manner?

thanks for help
klfi

[Fohdeesha]

no reason to transcode, AE and premiere pro automatically convert to 4:4:4 rgb 32bpc colorspace internally for all operations when the correct settings are chosen (maximum bit depth and maximum render quality in your sequence and render settings)

enjoy the advances of technology and just drop your BPAV folder right into premiere or ae, and have fun. You aren't missing out on any hidden quality by not transcoding to an intermediate codec, unless you're using a dinosaur NLE.

There's a time and place for intermediates, but simple editing such as that is not one of them!

[kolak]

Only thing which could improve your footage is special chroma upsampling, deblocking and some noise injection to improve gradients. NLEs use simple methods and have no special setting for this step.
This can be done eg. in avisynth. Will take some time- another step, but it could improve your source file and in the same time final graded footage. It's just a question if you have time and is it really worth- it will show up on some footage more than on other, depending on the nature of your video.