EX1 XDCAM EX vs. ProRes footage - added noise during conversion?
 

Hi all --

Don't have time to post frame captures (have to get to bed - beat), but wanted to get this out to see if anyone had seen what I'm seeing here. If necessary I can post frames tomorrow to illustrate.

I'm taking some footage from my EX-1, shot in 1080p24, -3dB gain, outside in bright sun. I walk in front of the camera with a color chart, wearing a dark gray shirt - I'm in the shadow where I'm standing.

When I view the footage rewrapped from Sony's XDCAM Transfer app, it's in the XDCAM EX codec, and it looks fine. However, after re-encoding to ProRes 422 HQ - it adds noise to the image - quite a bit of noise. It's noticeable when the movie it playing more than in a single frame, but you can tell the difference especially if you look in the shadows. It was obvious enough for me to stop what I was doing and examine the issue. If I use the non-HQ ProRes 422 codec as the compressor, the shadowed areas on my shirt are REALLY blocky and obvious even in a still frame - almost like a JPEG compression type thing.

Now, I'm not a codec expert or anything, but I thought that converting the footage to the ProRes codec wouldn't actually make the footage look WORSE, given that the data rate is higher than XDCAM EX as well as having more color information (4:2:2 vs. 4:2:0). However, what I'm seeing confuses me quite a bit. I'd think I should be seeing footage that looks so close to the same that I'd REALLY have to look to see a difference.

Now, if I output to uncompressed 8-bit, which I did just as a test, it comes out fine (and about 5x the file size). This leads me to believe that re-encoding to ProRes 422 HQ is more destructive than I had thought and could take quality out of the original long-GOP XDCAM EX-based footage...

For those who will ask: I'm re-encoding using Compressor, one of the several built-in ProRes settings, and I've fiddled with a variety of the options in the "Encoder" tab and the "Frame Controls" tab, including making sure that I'm not doing anything at all to the footage. In the end, I'm not resizing, retiming, deinterlacing, or anything else to the footage short of re-encoding it. I'm just amazed that I'm seeing output in a "more realtime" codec that actually looks worse...

Anyone else seen anything like this?

I have noticed that prores can be noisy. When I was looking at differences between recording in camera using xdcam and recording both prores and uncompressed via hdsdi I found that the prores version had more noise than I was expecting. So much so that in my chroma key tests I could get a cleaner key using the ex compressed footage than the prores footage. Of course the uncompressed gave the cleanest key.

Alister, are you talking about footage already recorded in camera and then taken in via sdi
through a capture card like a kona? Or are you talking about pulling in footage direct from
the camera to prorez?

Well, I guess it's good that I'm not doing something entirely wrong and that someone else is seeing the same thing, although I'd rather be wrong and corrected, with a perfect ProRes solution. I wonder if the new ProRes 4444 codec could be of any help here?

I guess the question I'd have for you, then, is what codec you ended up using for editing? In your opinion is ProRes not a viable option? XDCAM isn't my first choice for editing, especially since at some point you end up having to go to something like ProRes as an intermediate format for Color or After Effects, and using XDCAM might involve another recompression step - not good.

To see how much noise is being added try this test. In After Effects put the original XDCAM clip on a layer. Then add the Prores clip of the same file to another layer directly above it. Set that layers transfer mode to "Difference", only the pixels that are "changed" from the conversion will show up. Try this with different codecs!

Try stacking the same XDCAM clip and set the transfer mode of the one on top to "Difference", all you'll see is black.

You must make sure the clips are frame matched. If they are off even a frame, there will be a difference from pixel to pixel. But if you can frame match them (which is easy if you re-encode the exact clips without changing length), then only pixel to pixel differences will show, thus showing you just how much noise is added from the Prores conversion. This eliminates any noise in the original confusing things, because only the "difference" between the two will show. Hope this makes sense, it's hard to explain! But quick and easy to do.

Alister, are you talking about footage captured straight to prores from the SDI or XDCAM converted to prores? I have never had issues with prores being noisy while captured from the SDI and your remark about XDCAM producing cleaner keys over SDI to prores is what makes me think you were talking about XDCAM converted to prores rather than SDI to prores.

I'm probably missing something but is what is being said is that it's better to edit on an Uncompressed 8-bit timeline in FCP than Pro Res HQ?

Ryan, have you tried saving from from QT Pro as a ProRes file, or just using Compressor? The problem might be with Compressor.

This is an old thread and I haven't re visited this since last year, but at the time I was finding both converted and SDI captured prores material to be noisier than native EX footage. Now I have a NanoFlash I tend to use that at 100Mb/s and it's very clean.

The ProRes HQ footage from my Ki Pro is beautiful. There definitely isn't any added noise from that device.

Compared to what?

I think that it's 8 bit vs 10 bit that may be the visual difference here... It's been a tactic of high end post guys for some years to add noise to 8 bit footage to make it seem a bit deeper.

8 bit needs to do a fair bit of 'rounding' and what many of us see as being 'clean' footage is actually footage of objects that have more texture and variation than we have color precision to faithfully capture with 8 bits of color depth. Compression definitely accentuates

Compressed 10 bit footage can definitely seem more 'noisy' than 8... I've done some recording comparisons between 8 and 10 bit external devices fed via SDI...and I've seen what I believe you're seeing.

10 bit is still preferable for any compositing work...those minor tonal variations are a huge help in selling an effect in cases where using 8 bit footage would cause the grayscale to collapse and posterize...

My .02 anyway...

Ryan is going from XDCAM to ProRes 422 HQ, and then taking the same XDCAM and going to uncompressed 8-bit, which looks better to him.

If I understand what Tim is saying, is that Ryan is being confused by what he sees because the clean 8 bit uncompressed is actually footage of objects that have more texture and variation than there is color precision to faithfully capture with 8 bits of color depth, accentuated by the extra compression step added by ProRes, yet others are saying they see the noise going from 10 bit SDI to ProRes as well. That's a little perplexing.

Paul Frederick mentioned a clear way to prove his theory...

Compared to XDCAM EX recorded onto SxS.

One thing to consider is that the footage from an EX1 is border line with regard to getting any advantage from 10 bit recording as the noise level of the camera is greater than the 10 bit quantization level, so there is little to nothing to be gained by using 10 bit over 8 bit for capture with an EX1 (or any other camera with a 54db or worse noise figure).

Hey Ryan,
This has happened to me in exactly the same way and I stopped using ProRes because of the noise with the conversion using compressor. I don't know what is happening as PR supposed to be a clean compression but the extra noise has killed this for me.

How about giving Cineform a shot? It is a true intermediate codec that uses wavelet compression versus ProRes which uses the older DCT style compression.

I dont really use ProRes much, but I am always suspect of Compressor. I've even seen weird results that I dont like.

Alister, would it be beneficial to record 10bit from a camera like the new PMW EX 350?

The 350 is very clean so there may be a small benefit for recording 10 bit, but it will depend on many factors including the gamma you use and it's not going to be a big benefit, you'll probably still find camera noise the limiting factor in grading. You really want a noise figure better than 60db to really take advantage of 10 bit.

Maybe it’s more a question of being able to see more of what the camera actually sees with 10 bit over 8 bit than 10 bit actually making the imagine noisier which just doesn’t sound right. But I guess that’s the same thing you just said.


It’s more than preferable really. It’s basically a must! Reason why I couldn’t understand when Alister said he could pull a cleaner key from 8 bit 4:2:0 long-GOP over 10 bit 4:2:2 I-frame.
But it’s not only compositing. There are many benefits to be had in doing post in 10 bit.

Yep, I prefer Cineform over Prores. I think it has a higher quality.
But on a Mac Prores has some workflow advantages.

Yeah, I’m using STD3 on the EX1 when shooting green screen. The cine gammas do all sorts of funky things you want to avoid when shooting for green screen, like lifting the mid tones etc.

Well, but it’s not that 10 bit is worse than 8 bit. Whatever small gains it may produce, even if that’s the case, small gains is better than no gain. Besides, 10 bit seldom comes alone. If we were talking 10 bit 4:2:0 long-GOP over 8 bit 4:2:0 long-GOP I could see. But 10 bit is normally at least 4:2:2 and normally not long-GOP. Just leaving long-GOP land when you are talking about green screen and compositing in general is a Godsend. When you add 10 bit + better color resolution (4:2:2) + no long-GOP, it’s hard to say I would rather shoot HDV/XDCAM for green screen and compositing.

With an EX1 there is nothing to be gained from 10 bit, yes 4:2:2 will give an advantage, but you have to trade that off against the added complexity of capturing to 10 bit and the size of the files generated. It won't suit everyones workflow, but given the choice I too would prefer 4:2:2 at a higher it rate over EX 35Mb/s. That's why I use a NanoFlash at 100Mb/s. To get similar performance from ProRes you need to use HQ which generates much larger files, which slows down my end to end workflow and makes project backups larger. If you shoot progressive the difference between 4:2:0 and 4:2:2 is tiny, you are only gaining an increase in vertical chroma resolution, a decent keyer should easily be able to extrapolate that from a 4:2:0 progressive signal and most keyers don't just work off the chroma, it is a combination of chroma and luma. At the end of the day getting the lighting right is going to make a far greater difference than whether you are using native EX or something else.


Please don't mix HDV and XDCAM in the same sentence, they are very different, they use different mpeg profiles, different frame sizes and bit rates. It's like saying a Ferrari and a Ford are the same as they both run on petrol. The quality difference between HDV and XDCAM is significant.

When I did my original tests I found ProRes to be noisy, both converted files and direct HDSDi captured files exhibited more noise than the native EX1 or XDCAM HD files. I was using a Blackmagic card for capture, it is possible that this is not the best solution for ProRes capture. Certainly the uncompressed capture files were a lot cleaner. To date I have not seen anything yet to make me change that view, but I have not looked into it closely.

"Yep, I prefer Cineform over Prores. I think it has a higher quality.
But on a Mac Prores has some workflow advantages."


Well I'm in it for the higher quality, thats is what this whole thread is about after all.... I can only think of one advantage of Pro Res from a workflow standpoint and that is not a good enough reason to work with Pro Res for me. And I can think of some disadvantages of Pro Res as well, since a huge percentage of post facilities and VFX studios use Windows or Linux boxes.

I don't think in the EX1's case that it's a matter of what the camera "sees" being more intricately represented so much as what the camera is 'putting out' for signal. Alister's point is that many of these smaller cameras (I am an EX1 owner myself and really like that camera BTW) have a relatively high signal to noise ratio. It's part of the tradeoff in coming up with an inexpensive, small camera with a small sensor that puts out full HD resolution. I think Alister's point is that 10 bit is likely to represent this noise much more faithfully than 8...

I think there are some caveats involved here...I've seen 10 bit uncompressed from my EX1 and that stuff looks great (I'd have to rent hard drive space from the defense department to edit 10 bit uncompressed HD everyday, but...it looked very good). Once you start to have compression enter the picture, I think that has an impact on this issue as well.

Long GOP vs I frame for compositing depends on how good the codec is, and how well the compositing app handles it. It's not as if there are 'partial' frames that you can't composite of course...each frame is fully decoded and then the composite executes on that raster. I think when many of us think "Long GOP" we think of low bitrate files like you would see on a consumer DVD or in a web video. Long GOP alone doesn't determine quality level.

As far as 8 bit 4:2:0 vs 10 bit 4:2:2...and pulling a 'cleaner key'...

I think this needs to be placed in the context of lower end cameras with higher signal to noise ratios...as mentioned. If a 10 bit 4:2:2 clip has a compression scheme that accentuated some of that noise, you may end up seeing pixels of varying color on a greenscreen background resulting from noise that is now causing slight deviations in the chroma hue and luma info. In order to knock that background out, you need to create a 'wider bite' to make sure all the slight variations are included in the color you're 'knocking out' whereas 8 bit might have simplified that area, imaging what appears to be a more 'consistent' set of hue, saturation and luma values...in some cases 'less' is 'more'.

Also...4:2:0 vs 4:2:2 is a separate factor from 8 bit vs 10 bit. As far as 4:2:0 is concerned in general, I've puled keys from HDV. Is it ideal? Certainly not, but I think that many people somehow think that 4:2:0 keys like 4:1:1 (DV), and that really isn't true.

4:2:2 does up the density of color difference samples in the image and gives you an advantage over 4:2:0, but I think you also need bitrate to increase to realize any really serious gains as well. XDcam at 35 Mbit/s keys better than HDV because it's full raster (1920 vs 1440) and it has a 60% data rate increase. Both are 4:2:0, it's true, but at the end of the day, throwing more bits at it may not seem very 'sexy' or 'technical' but sometimes 'more' is simply...more.

In this discussion, the specifics are getting a little blended together...the actual range of formats and specifications you could be referring to within this realm of an '8 bit 4:2:0 Long GOP vs 10 bit 4:2:2 I-frame' discussion are many, and you have three factors that all change independently...and then you add different compression schemes on top of that with constant vs variable bitrate and DCT vs Wavelet transforms...

8 bit 4:2:0 Long GOP-

19 Mbit XDcam
25 Mbit HDV (not full raster)
35 Mbit XDcam
(also AVCHD falls in here at several bitrates)

8 bit 4:2:2 long GOP-

50 Mbit XDcam
100 Mbit MPEG2 (CD Nano)
140 Mbit MPEG2(CD Nano)
160 Mbit MPEG2 (CD Nano)

8 bit 4:2:2 I frame-

100 Mbit DVCProHD
135 Mbit HDcam
(these two tape formats are considered per manufacturer spec of course)
160 Mbit MPEG2 (CD Nano)
220 Mbit MPEG2 (CD Nano)

Also, several NLE manufacturers such as Avid, AJA, BlackMagic, Grass Valley have their own codecs that would fit in here...

10 bit 4:2:2 I-frame-

100 Mbit AVC Intra (Panasonic)
~36 Mbit ProRes 422 (Proxy)
~100 Mbit ProRes 422 (LT)
(and just 'ProRes' which has "significantly lower data rate than HQ" ...)
~220 Mbit ProRes 422 (HQ)

CineForm also belongs in here, with variable bitrates and several quality levels

Several different NLE codecs would also be in this list from Avid, AJA, BlackMagic, Grass Valley, etc.


...so which two are you comparing? The problem with discussions like these is that there are too many factors that get lumped together. The bottom line is that image specification is a game of inches. Lots of nuance here... Black and white conclusions are tougher to come by in the real world than they are on the sales brochures.

:-)