Web Analytics

Tested Sony FX9 ProRes RAW and Noise Reduction still works


Elvis Ripley
 

Over this holiday down time I rented an XDCA-FX9 raw back to do some Sony FX9 tests. I was surprised to find that the noise suppression as well as the distortion correction and shading compensation still worked on the raw out feed. 

Maybe this is due to the 6K to 4K downsample happening early in the chain and the noise reduction is all part of the downsample to a 4K bayer image. So unlike most cameras it isn't really a 'raw' file but a processed 16 bit bayer pattern image.

This is a great and welcome surprise. The 12 bit log raw files recorded by the Shogun 7 are a huge bump in quality over the internal XAVC and with a little noise reduction they are even better. This also changes my feelings on when I would use a Venice. I have been doing internal XAVC for normal jobs, SDI to Shogun 7 recording ProResHQ for jobs with a lot of post and Venice for the highest end. Now I think I would rather have 4000 ISO and variable ND with the FX9 if I can apply a bit of denoise in camera.

I wanted to do a write up about my experience because I haven't heard of anyone mentioning this anywhere but not many people are doing RAW from an FX9 because of the expensive back needed. Let me know if I am totally wrong or if you want to see anything else. 



My workflow...
To do these tests I underexposed plenty to make sure the noise was easily visible as part of the image.
Used Final Cut Pro to convert the ProRes RAW files to ProRes4444 S-Log3/S-Gamut3 files so they could be brought into Resolve for testing and ACES.

Here is an S-Log3/S-Gamut3 to sRGB version of the shot with no adjustment to show how low the exposure was.


Then in Resolve in an ACEScct timeline I massively boosted the offset and contrast for these images to make sure the differences made it through to CML. On the uncompressed files it is easier to see in motion than these stills.


This is a strip on the MacBook lid behind the ColorChecker. 


Looking at this color spectrum noise chart the differences are easy to see. This is Off->Low->Med->High on a loop.



And here is the noise present in the original image that isn't present in the High noise suppression setting.


https://f.io/17bGvb_c
Here is a link to a ProRes RAW clip where I just cycle through Low, Mid and High Noise Suppression settings as well as the same clip converted to ProRes 4444 with S-Log3 S-Gamut3 gamma and gamut if you want to check it out. 





--
Elvis Ripley
Director/DP Tulsa, OK


Pawel Achtel, ACS
 

Not directly related to Sony FX9, but I have experimented with noise reduction on my 9x7 camera and, similarly to your experience with Sony FX9, the noise reduction applied at the RAW level appears to be more effective and preserves significantly more shadow detail than noise reduction applied downstream on RGB demosaiced images. In my opinion this is the best place to apply noise reduction.

 

I’m using proprietary algorithm, but the principle should be the same. These steps can be applied in camera or in post process (RAW to RAW). The former gives you more convenience, the latter gives you more control and usually superior results because, apart from spatial, you can apply temporal noise reduction as well.

 

One thing to check is whether this noise suppression preserves detail in the shadows and how it compares with noise suppression applied downstream, like Neat Video.  Also, how does it behave with motion? I’m getting results superior to Neat Video (applied downstream).

 

On a side note, this doesn’t have anything to do with 6K -> 4K binning/averaging (which incidentally also helps noise suppression). These are additional RAW processing steps that can be applied at native Bayer pattern resolution too.

 

Kind Regards,

 

Pawel Achtel ACS B.Eng.(Hons) M.Sc.

“Sharp to the Edge”

 

ACHTEL PTY LIMITED, ABN 52 134 895 417

Website: www.achtel.com

Mobile: 040 747 2747 (overseas: +61 4 0747 2747)

Mail: PO BOX 557, Rockdale, NSW 2216, Australia

Address: RA 913 Coles Bay Rd., Coles Bay, TAS 7215, Australia

Location: S 42° 0'14.40"S, E 148°14'47.13"

Email: Pawel.Achtel@...

Facebook: facebook.com/PawelAchtel

Twitter: twitter.com/PawelAchtel

Skype: Pawel.Achtel

_._,_._,_


Mitch Gross
 

On Jan 1, 2021, at 9:58 PM, Elvis Ripley <4@...> wrote:

unlike most cameras it isn't really a 'raw' file but a processed 16 bit bayer pattern image.

Just a general FYI to you and anyone else reading this, pretty much every camera out there that outputs RAW processes the information in some way. All the marketing that signals are “straight off the sensor” is pretty much bull, and trust me it’s much, much better that way.


Mitch Gross
New York


Noel Sterrett
 

On 1/2/21 11:45 PM, Mitch Gross wrote:
... trust me it’s much, much better that way.
Due to limitations of size/weight/heat, processing capabilities inside a camera are far less than
outside. If by "better" you mean faster, perhaps, if you mean image quality, I have to disagree.


Elvis Ripley
 

For sure Mitch. I assumed that the image is always sweetened a bit but this is the first time I remember having control over the "raw" output.

Pawel it is doing a very good job with low noise suppression at taking the edge off and subtly cleaning the shadows. It isn't as pleasant as Neat and looks more like Resolve's temporal reduction where the grain is reduced but still has a little motion to it. This is the same Noise Suppression from the internal XAVC recordings so I am used to how it behaves. With the raw recording you are able to get way deeper into the shadow information than XAVC so the noise suppression artifacts can be more apparent. 
--
Elvis Ripley
Director/DP/Post
Tulsa, OK, USA
elvisripley.com