FS5: Comparison of Internal Recording on FS5 vs. RAW output to Odyssey 7Q+

[Doug Jensen]

In the months since I completed production of my 6-hour Sony FS5 Master Class training video I've had a lot of people write to ask me if I thought there were any advantages to recording the 4K RAW output of the FS5 to an external recorder vs. recording internally on SDXC cards. So, I decided to shoot some test footage one day with both at the same time and do a split-screen comparison. The camera only allows internal recording in HD resolution (not 4K) while 4K RAW is being output, so, in some ways, the test is also a comparison of HD vs. 4K. The camera was set for Picture Profile 7 (S-LOG2/S-Gamut) with the default settings. I exposed with the camera rated at ISO 1250 by exposing for white at 70%.

The 4K ProRes footage (on the left side of the screen) was lightly graded in Davinci Resolve 12.5 by compensating for exposure on one node and then applying Convergent Design's "SONY_EE_Slog2_LC709A" LUT on a second node. The XAVC HD footage (on the right) was also graded in Resolve using the same "SONY_EE_Slog2_LC709A" LUT, plus a few other tweaks to attempt a quick & dirty match with the corresponding ProRes file. I didn't spend a lot of time getting them to match perfectly so there are some remaining differences that I believe could easily be compensated for with a little more effort on most clips. No sharpening or noise reduction has been applied to any of the clips.


At 3:12 and 5:06 there are comparisons between internal recorded 4K XAVC vs. externally recorded ProRes 4K.


Lenses used were a Canon 200mm f/2.8 and an old Konica-Hexanon 28mm f/1.4.


FYI, outputting RAW from the FS5 requires several optional upgrades: First, you must purchase Sony's PXW-FS5 FS RAW Output Upgrade License which costs $500. Then you must have an external recorder that is capable of receiving the camera's RAW 4K output and either recording it as Cinema DNG files or various flavors of ProRes. I own a Convergent Design Odyssey 7Q+ which is ideally suited for capturing RAW output from both the FS5 and FS7. An Odyssey 7Q+ costs $1800 plus SSD media cards. But you also need the RAW BUNDLE license for the 7Q+ which adds another $995 to the cost. So, gearing up to record the RAW output of the FS5 requires an investment of at least $3500 over and above the cost of the camera itself. Is it worth it to shoot RAW? I'll leave that for other people to decide for themselves, my intent here was just provided some footage that may help with that decision.

 

[Arash Ayrom]

Hi Doug,

Interesting video that does raise some questions! Apart from the color differences, detail looked very similar.

With that in mind:

- Does the original file show more of a difference than the compressed Vimeo version? If there's any way to upload even a small portion of the "uncompressed" video somewhere to download, that would be awesome!

- Does the ProRes allow more gradeability than the internal XAVC? Again, a ProRes clip of the video would help answer that

- Is there an appreciable difference in noise characteristic and low light performance?

I did notice a small red square artifact popping up (I think it was over the rusty car fender close-up). Was this from the recording or the upload or something else?

Thanks for sharing this!

 

[Doug Jensen]

I will try to find some time to organize and post some samples for download, but that is more complcated than it sounds because the files are so huge.

- Does the original file show more of a difference than the compressed Vimeo version? If there's any way to upload even a small portion of the "uncompressed" video somewhere to download, that would be awesome!
!

No, the original files are very similar. To my eyes, they color on the ProRes looks more pleasing, but the ProRes is also nosier. So the choice is not easy.

- Does the ProRes allow more gradeability than the internal XAVC? Again, a ProRes clip of the video would help answer that

Not during "normal" grading. I did not try pushing either of them to unrealistic extremes to see which would break first because that would serve no purpose in the real world.

- Is there an appreciable difference in noise characteristic and low light performance?

Yes, I find the ProRes files to be noisier in the shadows.

I did notice a small red square artifact popping up (I think it was over the rusty car fender close-up). Was this from the recording or the upload or something else?

That flashing red artifact (seen at 4:42) is present in the original file captured on the Odyssey. I didn't notice it on any other clips.

It's interesting to note that the total data size of the project was:


4K RAW/ProRes: 351 GB
XAVC-L: 23 GB

[edited to fix earlier typo]

 

[Arash Ayrom]

Thank you for the replies, Doug!

And of course, life keeps us busy. No worries about uploading original clips

 

[Chadfish]

It's interesting to note that the total data size of the project was:

4K RAW/ProRes: 351 GB
XDCAM HD: 23 GB

Hi Doug, I just found this thread. I thought you recorded in XAVC for the internal. Does the RAW upgrade have XDCAM HD now too? Not clear how the XDCAM is implemented here.

Aside from my question, this really has me thinking if it's worth the Upgrade. Although I love the idea of having a 1500 nit 7" monitor to see what I'm shooting.

 

[Doug Jensen]

Hi Chad, thank you for catching that typo. You're right, I meant to say "XAVC-L" instead of XDCAM. That mistake was there for about 10 days and you were the first to say anything. Thanks.
I've gone back and changed the post to say:

4K RAW/ProRes: 351 GB
XAVC-L: 23 GB

You're right, the difference is slight. I guess someone would have to decide whether it is large enough to justify the extra cost and hassle or not of using an external recorder.

 

[DPStewart]

A decent test.
The FS5 is such an improved image over the Alpha series. Wow.

I just wanted to point out that some of these shots might not be really the best for comparison.
Let me explain.
First there is the basic optical principle invoked by close-ups. And that is that lower resolution always looks better when the subject fills more of the screen.
I even once used my crap garbage iPhone 4 on an event shoot and it WORKED. We set it as the tightest shot out of 4 cameras figuring "what the hell, we've already got 3 cameras on this, what have we got to lose." We actually got severl tight shots that cut against the Canon 7D's and hacked all-I frame GH2.
Now if the iPhone were set wider, we know what it would look like....crap. This also works when shooting in 720p on a good camera - if it's tigher it will look better. More pixels on the subject when it's tighter.
The mid and especially the wide shots on the FS5 here show a massive quality difference between the two methods. Not only in clarity and detail, but in color quality.

Next is movement. Very slow moving shots of objects can look identical on nearly every camera. Get the subjects moving a little more, or have flowing water of wind in the trees and we see how fast the compressed codec falls apart.
As an example that's crystal clear - Look at video from an iPhone 6s...when it's a static shot it looks AMAZING. But move the camera or much of the subject and the image immediately turns worthless.


The $5,000 of the FS5 in 1080p compressed is almost not even worth the cost. That image is nothing to write home about, and certainly doesn't compete against either compressed 4k or 1080p RAW.
But making that push to get the 7Q+ and record 4K in RAW!!?? Dayum! Now THAT is ready for full modern broadcast quality in ANY market.
$8K for a fully functioning, all the bells and whistles, with the 7-inch monitor, and everything you get with the FS5/7Q+ pairing, 4K RAW camera....that's a BARGAIN and I cannot recommend it strongly enough to anyone who already has an FS5.

 

[Doug Jensen]

It's good to hear your perspective and I can't really argue with most of it. In fact, I will also point out that these tests lack any human faces -- the very thing that we shoot more than anything else -- and care about more than anything else. But I've run out of family and friends who are willing to participate in such things, so it is what it is. Nevertheless, no matter how you slice it the difference are not huge between the two codecs, but that does not mean it's still not worth it to shoot RAW for some folks in some situations.

I'm trying to remain agnostic on the topic and not recommend or discourage any particular setting or workflow.

 

[NorBro]

I assumed people only upgraded for the 4K/120p bursts.

I think you would be really wasting your money if you upgraded for any noticeable visual quality improvement.

I love the FS5 because it's so small...that's what makes it fun.

 

[nyvz]

RAW should also bypass the processing that causes black pixels on high contrast edges, not to mention with XAVC you're stuck with worse black edge artifacts in HD but more muddy compression in UHD. But I have more experience with A7S/RII/SII's. For theatrical or broadcast it seems the RAW/prores is entirely worthwhile but for the web it may be more questionable.

 

[wschmid]

Dough, as I understand it you recorded the raw output of the camera to ProRes in the 7Q+ and found the ProRes to have more noise. Coming from the FS7 I know that noise reduction is an issue maybe. I assume that noise reduction has been applied in your test with the XAVC L internal recording, but has it be applied by the 7Q+ too? Because for raw every processing should be bypassed in the FS5 - and from that side I would expect the raw CDNG but maybe the ProRes to be noisier?

 

[Chadfish]

Hi Chad, thank you for catching that typo.

Oops! I guess you're so used to working with EX1s and EX3s that it just comes naturally to say XDCAM. I was on them for 7 years.

 

[Mitch Gross]

The camera applies noise reduction to the internal recording. This is not the case for the RAW output, and the Odyssey7Q+ does not apply noise reduction to the ProRes. It is easy to use one of the many noise reduction post programs to remove noise from the signal. By using a post program you can tailor the application to your taste.

The highly compressed internal recording can show artifacts on high frequency imagery. It would have been nice to see images with lots of complicated detail that was moving in random ways. Wide shots of trees blowing in the wind, wavelets of water on a lake or sea, congested traffic of cars and people on the sidewalks -- these are images that quickly challenge compression engines.

 

[wschmid]

The camera applies noise reduction to the internal recording. This is not the case for the RAW output, and the Odyssey7Q+ does not apply noise reduction to the ProRes. It is easy to use one of the many noise reduction post programs to remove noise from the signal. By using a post program you can tailor the application to your taste.

That was my understanding too! So if Dough compares not denoised "raw -> ProRes" with denoised FS5 internal recorded footage then the finding that "raw - > ProRes" has more noise is something that should have been expected.

And the second issue with the test is that is does not address the reserves one will have in the postproduction for corrections and adjustments. The internal 8bit 420 footage will become blown up much earlier, compared with CDNG 12bit or even 10bit 422 ProRes.

And very true - the FS5 internal hardware encoder would be challenged much more by motives like cornfield or trees in the wind.

 

[Nirv]

And very true - the FS5 internal hardware encoder would be challenged much more by motives like cornfield or trees in the wind.

But XAVC isn't unduly troubled by motion, it is incredibly robust where motion is concerned.

I would love someone to prove a ProRes recorder can out perform the internal XAVC-I Codecs in HD and UHD of the FS7. If the a ProRes recorder can out perform XAVC-I in what circumstances is it better and is the difference so vast it's worth the extra data load to the project?

I'd like to see UHD out of the HDMI recorded on the best Prores datarate not via the RAW box from the FS7 because we know that's naff.

 

[Doug Jensen]

The camera applies noise reduction to the internal recording. This is not the case for the RAW output, and the Odyssey7Q+ does not apply noise reduction to the ProRes. It is easy to use one of the many noise reduction post programs to remove noise from the signal. By using a post program you can tailor the application to your taste.

The highly compressed internal recording can show artifacts on high frequency imagery. It would have been nice to see images with lots of complicated detail that was moving in random ways. Wide shots of trees blowing in the wind, wavelets of water on a lake or sea, congested traffic of cars and people on the sidewalks -- these are images that quickly challenge compression engines.

Well that's always the problem with doing testing, isn't it? Someone always says you didn't test the right way, under the right circumstances, and with the perfect combination of settings that would show differences. I don't know why I would waste by time doing more testing, since I don't have a horse in the race and really don't care one way or the other, but nevertheless, I will try to find some time this week to shoot some new test shots that have plenty of action.

 

[Doug Jensen]

That was my understanding too! So if Dough compares not denoised "raw -> ProRes" with denoised FS5 internal recorded footage then the finding that "raw - > ProRes" has more noise is something that should have been expected.

Okay, now you just added another $1000 to the cost of shooting RAW because someone can't do noise reduction with the free version of Resolve. So, now the cost of shooting RAW is $4800 instead of $3800. Plus adding any NR will triple or quadruple your rendering times. So, better budget more time for post if you shoot RAW. I'm sure the counter argument will that there are less expensive ways of doing NR, but that is just making excuses and scrambling for workarounds to a problem that doesn't exist with the internal footage. If someone does their grading in Resolve (as should anyone who is going to shoot RAW or S-LOG seriously) then you should use the built-in NR for best results. And that will cost someone $1000. I just happen to own two full licenses for Resolve Studio so it's not a big deal for me to add some NR in post, but that misses the whole point of my testing and I won't do it. BTW, I never add NR to RAW from my FS55. I can't even remember the last time I used any NR in Resolve, so to suddenly have to start thinking in those terms for FS5 RAW is a step backwards.

And the second issue with the test is that is does not address the reserves one will have in the postproduction for corrections and adjustments. The internal 8bit 420 footage will become blown up much earlier, compared with CDNG 12bit or even 10bit 422 ProRes.

As I said in an earlier post I'm going to try to do some more testing this week, so tell me what you want to see. What specifically can I shoot that will meet YOUR requirements to show this phenomenon where one codec "blows up" sooner than the other? And please don't tell me to shoot 3-stops underexposed, or directly into the sun, or shake the camera violently from side to side, or some other way to purposesly shoot bad footage. I don't shoot bad footage so that would not be a realistic test. Give a real-world situation where I am virtually guaranteed to see the difference whnen shooting properly exposed, realistic footage, and I'll do my best to create those conditions.

And very true - the FS5 internal hardware encoder would be challenged much more by motives like cornfield or trees in the wind.

We'll see. I have not noticed any issues whatsoever when shooting action with the FS5 in the past, so it is hard to imagine how it is going to look better with RAW. Is it possible to fix something that isn't broken already?

 

[jc01108]

Okay, now you just added another $1000 to the cost of shooting RAW because someone can't do noise reduction with the free version of Resolve. So, now the cost of shooting RAW is $4800 instead of $3800. Plus adding any NR will triple or quadruple your rendering times. So, better budget more time for post if you shoot RAW. I'm sure the counter argument will that there are less expensive ways of doing NR, but that is just making excuses and scrambling for workarounds to a problem that doesn't exist with the internal footage. If someone does their grading in Resolve (as should anyone who is going to shoot RAW or S-LOG seriously) then you should use the built-in NR for best results. And that will cost someone $1000. I just happen to own two full licenses for Resolve Studio so it's not a big deal for me to add some NR in post, but that misses the whole point of my testing and I won't do it. BTW, I never add NR to RAW from my FS55. I can't even remember the last time I used any NR in Resolve, so to suddenly have to start thinking in those terms for FS5 RAW is a step backwards.



As I said in an earlier post I'm going to try to do some more testing this week, so tell me what you want to see. What specifically can I shoot that will meet YOUR requirements to show this phenomenon where one codec "blows up" sooner than the other? And please don't tell me to shoot 3-stops underexposed, or directly into the sun, or shake the camera violently from side to side, or some other way to purposesly shoot bad footage. I don't shoot bad footage so that would not be a realistic test. Give a real-world situation where I am virtually guaranteed to see the difference whnen shooting properly exposed, realistic footage, and I'll do my best to create those conditions.




We'll see. I have not noticed any issues whatsoever when shooting action with the FS5 in the past, so it is hard to imagine how it is going to look better with RAW. Is it possible to fix something that isn't broken already?

Nice comparison footage Doug. Just lovely in addition to being informative!

Raw for the FS5 seems like a lot of work and cost for something that will marginally improve the footage. For my clients, they would never see the difference and for the cost, I would much rather spend the money on a new lens or sell the FS5, which I like for its form factor as much as anything else, add the money I would spend on the Raw upgrade and hardware and buy an FS7 which supplies so much more in terms of codecs and features than the FS5. I am not planning on doing that since I find the small form factor a blessing for doc work and have not been hampered in any real way by its supposed short comings. Still love to see comparison footage especially from Doug who shoots such lovely footage.

 

[morgan_moore]

Doug

It seems (to me) that to some extent you miss the purpose of raw whose strengths (if there are any) would only show when the footage is given an extreme manipulation or some other codec stress. It is not only operator error that stresses a codec.

-frizzy hair on green screen
-matching two shots either shot in different locations or during seriously changing light
(eg. inteview of two CEOs, one in LA, on in NY)
(long interview as sun drifts in and out changing ambient colour temp)
-tracking shot (eg flouro kitchen steadicam to tungsten eating area)
-Heavy 'brand colour' look
eg all LLoyds bank images are green - but a rebrand might require them all to go pink!
-huge DR - room with sea view
-fashion grey backdrops with subtle tonal ranges
-complex architectural scenes, that may need tombstone correction.
-complex architectural scenes
-fibre, fabric

Obviously you would not be able to recreate these shots during a no budget test.

My standard 'stress test' is a grey curved surface, sidlit with tungsten, shot with the camera set on 5.5k, how well does the footage correct?

If I had acces to raw I would only use it when I elected that I needed it. Eg greenscreen. Something that was by definition going to stress the codec.

S

 

[nyvz]

The idea that there is only marginal difference between 50/100Mbps XAVC and 200/800Mbps Prores seems like a matter of different perspectives. If you shoot for the big screen or for workflows that are used to dealing with images from Arri/RED/etc you will notice the difference in an instant, but if you only ever screen in HD or on the web and are used to the limitations of highly compressed formats then you may not see a difference. Compression and many of the intricacies of image quality is highly subjective. Yes high compression is less noisy but that's because it has NR and the codec naturally smooths out high frequency and especially random detail to make it easier to compress, but then you also get muddiness that cannot be undone. If your footage is going through heavy compression for delivery anyway that will mud it up and remove noise then there is certainly a point of diminishing returns in trying to retain very high frequency detail or noise texture on set.

You see this a lot in the RED world where some people swear by 7:1 compression and lower and others go for 12:1 or 18:1 and think it looks just fine. The reality is it is a sliding scale that depends highly on how hard you push your footage in post, the type of scene content and motion you shoot, and how you are monitoring.

In my experience you have to be a bit careful expecting too much from highly compressed footage but it is more than enough and in some ways absolutely necessary for a lot of styles of shooting, such as corporate or documentary or low budget web commercial etc.