ML RAW and In Camera N/R?

BobbyMurcerFan

Well-known member
I am really falling in love with the full frame look, but I have a question about using ML RAW on the 5DM3.

From my experience, Canon has excellent in camera noise reduction. Does using ML RAW bypass the in camera N/R?

BTW, I have NeatVideo but have found good in camera N/R to be superior in most cases.

THANKS MUCH!
 
Yes there is camera playback now.... and you would'nt want the in camara noise reduction if you shoot RAW, you will use ACR for that and get much better results!!
 
I know it sounds like wanting your cake and eating it too. But Canon's color science second to none. If it's being completely bypassed that's too bad. FWIU, Canon's DSLR's know when a portrait, landscape, sunrise/set or foliage is being shot and adjusts the color science accordingly. Trying to accomplish the same in Resolve would be very difficult.
 
I don't understand what you are lamenting about raw video on the 5D3. Do you also lament the ability to shoot raw stills vs. jpegs straight out of the camera?

In my opinion, you are not losing the Canon look when shooting raw. Just ask many wedding and portrait photographers who think that Canon skin tones when shooting raw stills are second to none. You are not losing that when shooting raw video. Canon's color science has more to do with the color filters on the sensor, and the sensor itself rather than the digic chip that makes jpegs and h.264. Are the picture styles the same as raw? No they are not. As far as I am aware, no photo, video, or cinema camera has a predefined built-in look when shooting in raw. That is one of the reasons for shooting raw in the first place. You can decide the look after the shoot in post to a greater degree.
 
Except what can be relatively manageable in Photoshop when working on a still can become much more challenging in Pr, Ae, Sg, MC, Symph, Magic Bullet, Resolve or Baselight when images start to move. And I kind of doubt that a DNG converter does as good a job on DNG files as a Canon RAW converter does on Canon RAW files.

And we have no idea how much image processing Canon slips into its RAW info.

Not saying RAW out of the 5DM3 isn't a great accomplishment. It certainly is.
 
High quality noise reduction is very expensive to compute, especially in order to preserve detail (microcontrast) and resolution. The best debayering solution right now for 5D3 RAW is Adobe Camera Raw (ACR) and After Effects. It denoises and debayers, corrects for lenses, recovers highlights/shadows, etc. The challenge is processing time. Resolve 10 is fast to debayer (~18fps vs 1-3 for ACR+AE on my 4-core MBP laptop), however the quality isn't as good, and getting the colors correct is challenging with Resolve (perhaps easier for a Resolve expert and perhaps using a proper 3D LUT). ACR with all default settings (and NR as needed) looks pretty good. After debayering and NR, more color work can be done in 10 or more bit compressed codecs (ProRes HQ and DNxHD are the most popular).

On the topic of noise reduction: reading research papers shows that the most advanced debayer algorithms also perform noise reduction at the same time (part of the statistical analysis). Some even perform Super Resolution (extracting real resolution increases from lower resolution data using aliasing information across frames (which will happen with a Bayer system)). The 5D3 RAW Bayer array is only 1920x1080. That's not enough resolution to fully support 1920x1080 RGB (debayering can't ever get back to 100% original resolution, except with Super Resolution and motion analysis (very expensive to compute), which can actually exceed the single frame original resolution: http://en.wikipedia.org/wiki/Superresolution). BMD, for example, uses 2.5K to provide full 1920x1080 after debayering.

Someday we'll have 10+ bit 422 XAVC (or similar) options to further reduce disk space and maintain or even improve image quality at lower bit rates. IPB codecs, which are very efficient, can look as good as ALL-I for low motion, but current implementations break up with fast motion. There's no reason the codecs can't switch to ALL-I during fast motion and provide the best of both worlds (no reason other than CPU/memory constraints in cameras, however on desktop computers there's nothing stopping innovation here).
 
jcs, Thank you for the excellent post. My only contention is that what's expensive in computing power for a program running on top of an OS is much less intensive when using a processor designed specifically for the task.

Also cameras can add analog gain, which truly helps with low light images. If Canon adds analog gain to its RAW files, IHNI. But I wouldn't be surprised.
 
ML RAW is better than internal in absolutely every IQ measure, including color. Yes, the Canon color science is pretty good. But ML-RAW is even better.

From here:
The following screenshots are with a Canon 50mm f/1.4 and with a Leitz Elmarit-R 90mm f/2.8, on a 5D2+RAW, processed with ACR. The small squares are the "correct" tones.

attachment.php

attachment.php


For comparison, these are the results with some common picture profiles for Canon DSLRs.
 
Samuel, I really do appreciate the post and all of the other helpful information that you post. But a color chart is not going to test to see how the camera renders skin tones slightly differently when it sees a face.

Color science is very complicated. And Canon spent a lot of effort to get theirs down so it would mimic film.

I realize that ML-RAW is a great achievement and that if I get 5DM3, no one will be forcing me to use it. I think the benefits outweigh the drawbacks. But I continually am impressed with footage from the 1DC and 1DX. And I'm thinking color science may be an important part of the answer.
 
Filmconvert can be used to map colors from the 5D3 to match film stock. While complicated, the practical implementation is likely a 3D LUT (which is simple). Analog gain will be related to ISO and fixed (if it changes at all- I believe it's fixed analog gain at the native ISOs and digital gain at intermediate ISOs). If gain and color were changing when a face was detected, background scene colors would change as faces enter/exit scenes and that does not happen.

1DC/1DX 1080p FF is pretty much the same as 5D3 1080p in terms of real resolution and color.

The best color science is currently found with ARRI (Alexa, Amira). They figured this out from their digital film scanning experience. 5D3 RAW + ACR is like a baby Alexa.
 
Last edited:
Samuel, I really do appreciate the post and all of the other helpful information that you post. But a color chart is not going to test to see how the camera renders skin tones slightly differently when it sees a face.

Color science is very complicated. And Canon spent a lot of effort to get theirs down so it would mimic film.

I realize that ML-RAW is a great achievement and that if I get 5DM3, no one will be forcing me to use it. I think the benefits outweigh the drawbacks. But I continually am impressed with footage from the 1DC and 1DX. And I'm thinking color science may be an important part of the answer.

That may be, but then they poo pooed up royaly… I worked with a beauty photographer in London for a while doing his retouching, and he was cursing his Canon 1Ds cameras because they treat skin tones awfully, and they really do. I haven't owned a single Canon camera which I like the skintones on, straight out of the camera.
Dunno why that is, but it's just not good enough.
 
The Colour rendition of Canon camera's has always been a significant plus as far as popular opinion... this is the first time I have ever seen a post stating otherwise. I cant understand how he could not get the colour he desired... very odd indeed. I could name literally hundreads of fashion and beauty photographers that use Canon and I have never read or heard any negativity regarding skin tones, it doesn't make sense at all.
 
I know there's much more to skin tones than some color patchesk in a checker board. But it's a very good indicator. In any case: skin tones in RAW video are very, very, very nice. To me, better than what you can get in H.264 with any picture style (and that includes Flaat).
 
Samuel, thanks for the input. I have to say that if one wants full frame, then the 5DM3 is the best choice out there.
 
The Colour rendition of Canon camera's has always been a significant plus as far as popular opinion... this is the first time I have ever seen a post stating otherwise. I cant understand how he could not get the colour he desired... very odd indeed. I could name literally hundreads of fashion and beauty photographers that use Canon and I have never read or heard any negativity regarding skin tones, it doesn't make sense at all.

Probably semi-pro / consumer thing...

The Canons are a bit too red / magenta to be honest. That's something I hear from pretty much all the colleagues I speak to... the thing is, "a bit too red" is not something we wanna bother spending time going through official channels to complain about and so on.

If I could have it... I'd want the skin tones of a medium format camera like the Phase One. But I know that is asking a lot.
 
Back
Top