FS5: FS5 Internal Recording vs. ROdyssey 7Q+ -- PART 2

[Doug Jensen]

Here's another (and final) round of testing to compare the FS5 internal vs. RAW output to the Odyssey. Some people said they wanted to see more gradients, motion, action, and fine detail, so this is the best I could come up with yesterday afternoon. (Sorry Nirv, no blue skies yesterday).

I also took the opportunity to compare some other format options as well:

Internal HD vs. external 4K via the camera's RAW output
Internal 4K vs. external HD via the camera's SDI output
Internal 4K vs. external 4K via the camera's HDMI output
Internal 4K vs. external HD via the camera's HDMI output
Internal HD vs. external HD via the camera's SDI output

If you want details about how I did the tests you can find more info on the Vimeo page.

 

[Arash Ayrom]

Link in case it's not showing up for you (it didn't for me):

https://vimeo.com/183839655

 

[Publimix]

Nice work. Sadly my eye is not trained enough to see much difference. I think I need to see all together on ten or twelve similar screens to see the striking differences.

 

[ozmorphasis]

Nice. Thanks for posting!

Some observations.

!. The raw to prores is prettier. Hard to tell how much of that is because it is more saturated. Why is it more saturated would be my first question for such a test.
2. The raw is much noisier in the shadows.

 

[diegocervo]

what impresses me more is XAVC actually

 

[Nirv]

Good Job Doug.

The internal noise washing and XAVC codec make a very strong case for web delivery.

 

[morgan_moore]

I didn't see anything that made me think.. yep I want to quadruple my data rate

On a casual watch there is not much difference to me.

One of my bugbears with the FS is the fringes on specular highlights eg boat chrome rails.. and the onboard codec looked better!

--

Doug is this Slog?

--

We are still not seeing any major bending of the codec in post afaik. Doug you are too nice to your cameras

This film I took all the colour out and added a big Scurve for a black and white print look and a few huge secondary 'burns'. You can see the FS7 sky tonality is completely breaking (2nd shot) .. would raw have helped?
https://vimeo.com/165058350

S

 

[markfpv]

Thanks for sharing Doug. Very interesting to me that the internal L codec holds up so well in shots with lots of motion. As FS7 owner, I shoot 90% using XAVC-I, but I feel your tests conclude that - for web delivery at least - saving a bit of card space by using the Long GOP codec would be just fine.

 

[morgan_moore]

Doug.

If you dont mind Ive given one of your shots a codec crunching grade. Obviously there have been many steps of recompression so quality should not be judged as the higher bitrate/res should be graded before output/downscale - although the XAVC sky is begining to fall apart - really Im just showing the sort of hoops I would expect footage to be thrown through..

Client 'can you make the shot more sunset'
Colour 'certainly'

https://vimeo.com/184001242 PWD crunch

 

[markfpv]

!. The raw to prores is prettier. Hard to tell how much of that is because it is more saturated. Why is it more saturated would be my first question for such a test.
2. The raw is much noisier in the shadows.

After downloading Doug's file from Vimeo and watching again, it seems more apparent to me the noise in the shadows of the raw shots is some sort of green channel thing. Pause the scene at :05:28 - 05:39 and look at the difference at the shadowed background that's out of focus behind the flowers in the foreground. That where the difference is most pronounced imo. Most of the green colors in the raw shots do seem to have more saturation.

 

[morgan_moore]

Client "Can you punch in on that plane a bit and blue up the sky?"
Colour "sorry that is a bit much for the codec it will look a little shaggy"

https://vimeo.com/184004269 pwd crunch

 

[Doug Jensen]

Doug.

If you dont mind Ive given one of your shots a codec crunching grade.

I don't mind at all. Do whatever you want to do with any of it for testing purposes.

FYI, I'm rendering a new version from Resolve right now (going back to the original unadulterated footage) that adds a final node on top of the orignal grade that applies a pretty heavy (and somewhat ugly) extreme look. I'll post it later in the day.

 

[Demistate]

Nice. Thanks for posting!

Some observations.

!. The raw to prores is prettier. Hard to tell how much of that is because it is more saturated. Why is it more saturated would be my first question for such a test.
2. The raw is much noisier in the shadows.

This is because the way Sony Processes their video, like temporal De-noising, lens corrections, and even their linear to log conversion does not get sent over the wire to the recorders.

On the Atmos Shogun you can actually select between Slog2 and Slog3 encoding into your prores.

The processing is different.

 

[Doug Jensen]

I'm adding some de-noising with Resolve to the ProRes footage (not the XAVC files) because it is so obvious it needs it. The RAW stuff is especially ugly when an extreme grade is applied -- and even 11 points of de-noising doesn't help much. Unfortunately, the de-noising has dropped my rendering speed by approximately 450% so maybe I can post the next version in a week or two.

 

[wschmid]

Doug.

If you dont mind Ive given one of your shots a codec crunching grade. Obviously there have been many steps of recompression so quality should not be judged as the higher bitrate/res should be graded before output/downscale - although the XAVC sky is begining to fall apart - really Im just showing the sort of hoops I would expect footage to be thrown through..

Client 'can you make the shot more sunset'
Colour 'certainly'

https://vimeo.com/184001242 PWD crunch

That is one comparision that tries to take into account that we have to grade the footage - and I think that is well done from that side. To my opinion it is quite clear that the ProRes footage seems to show also blocking, but less then the XAVC-L part.

BUT you did not have the original footage at all, even if you have downloaded the "original" footage from vimeo then you do that based on the rendered output but not the original footage. This is a limitation to the test and should have been done better based on the original footage.

I'm adding some de-noising with Resolve to the ProRes footage (not the XAVC files) because it is so obvious it needs it.

That is why I stated that this must be involved, even if we tend to test the total production chain then (and yes that makes more sense to do so).

 

[Chadfish]

Man this really shows just how little the external RAW recorder adds visually. I still want an Inferno for some reason! Something about the 1,500 nit 7" screen is calling to me. Why can't there be a giant difference in quality LOL?

 

[morgan_moore]

BUT you did not have the original footage at all, even if you have downloaded the "original" footage from vimeo then you do that based on the rendered output but not the original footage. This is a limitation to the test and should have been done better based on the original footage.

I did make note of that point in the post. "Obviously there have been many steps of recompression so quality should not be judged as the higher bitrate/res should be graded before output/downscale"

I was just making the point that such a hard grade (with the original material) should show up the differences.. if there are any.

S

 

[Doug Jensen]

I have uploaded a new version of the PART 2 test that now has 11.0 points of spatial Noise Reduction applied to the RAW/ProRes files (not the XAVC files) in Resolve 12 Studio. I think this will make some people happy who felt I was not giving the RAW/ProRes files a fair shake when I didn't apply any NR. Obviously NR is mandatory with the RAW/ProRes file. But of course, there is a huge cost in time to render with NR. The inernal XAVC files render at about 22 fps with a light grade in Resolve, but the RAW/ProRes files with NR render at about 1.5 fps. That is 1400% slower.

 

[morgan_moore]

Obviously NR is mandatory with the RAW/ProRes file.

This opens up a different can of worms. I would suggest not 'obvious'

-realistic dynamic range
-ETTR

In working with early stills raw cameras (where there was no 'lightroom style' processing - which does a bunch of stuff 'under the hood') it became obvious that one needed a grasp of the concepts of ETTR and 'realistic dynamic range'

Simply if one lit ones scene (remember lighting?) bright enough and with a thin enough DR (aka got ones lighting ratios correct.. thinner than film) noise reduction was not required.

NR was an emergency tool in the box, so when shooting raw maybe at a nighttime theatre performance (with no control of the lights) then one would be thinking 'Im gonna have to dig into the blacks here' and probably grab some noise reduction'.. but with well lit scenes no NR would be required.

Clearly for filming 'people out boating' or 'herons on the lake' beyond picking magic hour, or getting a gigawatt lights rig, the operator has little control over scene dynamic range (beyond allowing elements of the image to clip at one end or the other)

I have still never seen a test shoot that fully explores ETTR.

S

a side note on long render times and using NR in Dresolve. One can apply a global node with NR/Sharpening and leave it off while colouring (as it can severely clog playback) and only switch it on for the final render. If that render takes a long time it may not matter.. if you are asleep/at dinner whatever. Of course for fast turnarounds render is an issue.

 

[sidd]

My mind is blown.

How on earth a 4k frame, downsampled to a 1080p timeline, have noticeably less detail than a 1080p frame.
The markings on the oar, the straps on the boat, the lady's skin - everything has less detail on the 4k raw->prores side of the frame.


Something along the chain is not right - starting from the way the RAW is output by the camera/interpreted-encoded in Odyssey to the final processing/render in Resolve. Maybe the NR is to blame.

Whatever it is, I am very very surprised.