Results of testing various 5D3 RAW workflows

J

jcs

Veteran
I've been testing many workflows for 5D3 RAW. I can edit DNG's in real-time in Resolve 10 on my late 2013 4-Core MBP. Resolve can write out to ProRes HQ at around 18-20fps, so it could be a fast batch processor. Unfortunately, even with sharpening turned up, it can't really come close to ACR. A/B'ing clips shows ACR pulls out way more real detail, part of which is possible (apparently) from denoising during debayer (vs. after). The colors also look more 'right' with little effort in ACR vs any other tool I have tested so far. Until something better comes along, it looks like ACR+AE => [ProRes HQ(185) | DNxHD (175)] is the way to go for the best 5D3 RAW.

Adobe CC can edit and display RAW in real-time, however there's a pink cast over the entire frame and no access to ACR to make any adjustments to the RAW Bayer data.

mlrawviewer is an awesome tool to preview RAW clips in real-time (with sound if using MLV). Once installed you can double click MLV files and play them in real time. The AMaZE debayer looks very good compared to ACR (AHD variant?), even without denoising. However the color processing isn't there yet, and its direct save to ProRes HQ is close to the same speed as ACR+AE (MP enabled). The developer (Andy) is actively adding new features (he just added brightness and basic color balance controls), so it may be become more useful for converting to ProRes later on.

On my 12-Core MacPro with Render Multiple Frames enabled (MP) in AE, 5D3 RAW can be processed into DNxHD (175) at around 1/4 real-time (full CPU utilization). Was hoping to find a faster solution, however nothing else comes close to ACR+AE for color and image quality.
 
Thanks for sharing.... I have also chosen that route for myself. And a nice heads up regarding the processing on a 12-core.
I had a dream that a more powerful computer would drasticly speed up ny prores renders but I guess now the price cant justify the upgrade :-(
 
Yes, but isn't the problem with ACR+AE that if you make any kind of serious corrections including working with the highlight retention slider you cause flickering and other anomolies in the output?
Thus it remains an unworkable solution for all but the most basic of gamma corrrections?
 
That is a problem that you can overcome by using the 2nd tab in ACR and playing with curves to draw back your highlights, although I didnt get any flickering on my last project using all of the sliders, but that was "in my opinon" because the light temperature never changed. The same flickering can happen when you do timelapses and that is generally caused by using auto white balance. I might take a cutom white balance shot and see if this helps completely solve the problem when shooting in changing light.
 
I have also come to the same conclusion as JCS... I will fail to put it as eloquently but for me ACR gives far better results in terms of sharpness and noise reduction. As I said, I could not tell you the science behind my findings, I just know that when I play with ACR I get some much more from the RAW files than in resolve. It takes a lot longer to process the files... but its worth the wait IMO.
 
yoclay said:
Yes, but isn't the problem with ACR+AE that if you make any kind of serious corrections including working with the highlight retention slider you cause flickering and other anomolies in the output?
Thus it remains an unworkable solution for all but the most basic of gamma corrrections?
Click to expand...

Never seen anything like that because of ACR and I mess around a lot with both the whites and the highlights.

The only time I saw flickering was because of one of those annoying settings in ML where you can set something to auto...
 
Legrevedotcom.. there is definitely an issue, its been discussed at length on the Magic Lantern forums.. It happens when you play with the shadow, black white and highlight sliders in ACR...BUT.. in my opinion, I think its something to do with changing light conditions. I have had it when I was taking some test shots outside.. the flickering occurred only when the light conditions changed...that's why I was wondering whether it had something to do with the autoWB similar to shooting timelapses. On shoots where the light temperature is constant, the flickering disappears..
 
If you shoot talking heads, static shots, uniformly lit shots, or anything like this, tonal distribution in the image doesn't change, so you get no flickering. But ACR shadows, highlights, etc sliders are adapting their range to the specific frame and may introduce flickering if your shot pans across varying lighting conditions.

I am not sure ACR actually denoises during debayer. What you see probably has more to do with its camera specific noise profiles. You do need ISO metadata in your DNG frames to get the most of these.
 
Well... honestly, why would anyone ever use Auto anything.... its raw footage.

This was shot with raw and put through ACR when imported into AE as a sequence. I can guarantee that I moved both shadow, blacks, whites and highlights in every shot :)

 
jcs said:
A/B'ing clips shows ACR pulls out way more real detail, part of which is possible (apparently) from denoising during debayer (vs. after). The colors also look more 'right' with little effort in ACR vs any other tool I have tested so far.
Click to expand...

Would you mind posting those A/B frame grabs?
 
There is flickering in some cases- I'm studying it to make sure I understand when it can happen so it can be avoided. Worst case- just do debayer+NR in ACR and do grading in AE with a 16- or 32-bit project before rendering out to 10-bit ProRes or DNxHD (175+Mbit/s).

Minor flicker example: https://vimeo.com/88478272 (link for iOS etc.)

Playing back on OSX I can see pan judder: that's the browser/OSX (Chrome was slightly worse than Safari). Playing back the link on my iPhone 5S is butter smooth during the pans.
 
Last edited:
Steve Kahn said:
Would you mind posting those A/B frame grabs?
Click to expand...
ACR
ACR1.jpg
Resolve 10
R10.jpg

Uncompressed PNGs:
http://www.brightland.com/t/ACR1.png
http://www.brightland.com/t/R10.png
[EDIT: was able to improve microcontrast with different settings to closely match ACR in detail: the parts of the shirt pattern that were completely disappearing in Resolve are now showing up]


Overall colors are also a challenge with R10- easy with ACR. Too bad R10 can't match ACR as it is so much faster (~18fps vs 1-3fps on 4-core laptop).

Note also that ACR can do lens corrections (which I was testing in this example- the ACR frame is undistorted (slight shift if you A/B the PNG's with Photoshop layers, browser tabs, etc.).
 
Last edited:
I think you are misinterpreting contrast as being sharper. There is no difference in "real detail" between those shots. Resolve shot looks quite natural.
 
Last edited:
Microcontrast is detail- many parts of the shirt pattern disappear into mush with Resolve. If you A/B the uncompressed PNGs by opening both you can toggle browser tabs (and zoom in) and the difference is more clear. Playing back real-time Resolve was clearly lower detail.

Folks with ACR can do the same test themselves- Resolve 10 Lite is free. Another test would be to get a good color match in the full version of Resolve along with Resolve's noise reduction to compare with ACR.

If someone who's an expert with Resolve can show that it matches or exceeds ACR, I'd love to see the results and learn the steps necessary (including purchasing the full version).
 
Last edited:
Resolve does OK compared to ACR for debayer in luminance, however there were issues with magenta artifacts (chroma aliasing) for low light:
http://blog.lightandtime.co.uk/post...mera-raw-vs-davinci-resolve-for-magic-lantern (he makes a lot of good points as well).

This was stunning (done with Resolve 10 a while ago): http://www.magiclantern.fm/forum/index.php?topic=8932.0
It's all bright light (no NR needed or used). He archived to ProRes 4444 vs HQ.

Resolve is so much faster than ACR (around 10x), provides excellent grading tools (and very cool tracker), is possible to do edits in real-time on DNGs: experimenting more with it. If possible to get close to ACR and Neat Video can be used for final noise clean up, this will be a much faster workflow.
 
After spending more time with Resolve, I can get it looking pretty good relative to ACR (debayer quality looks similar to ACR; haven't matched ACR default colors yet (which look really good)). On this MBP 4-Core (2013), it's possible to edit DNGs directly from an external USB3 drive in real-time (barely) (WD My Passport Ultra 1TB (5400RPMs, about 115MB/s read speed), around $80). Working off the internal SSD is easily real-time, as would an external SSD or faster USB3 or Thunderbolt disk device (7200RPM, RAID, etc.). USB3 is plenty fast and much cheaper than Thunderbolt.

Something very cool: Neat Video and Film Convert have OFX plugins which work with Resolve and GPU acceleration ($199 each). Film Convert with this setup runs 14fps (OpenCL GPU). Neat Video (demo, smaller processing area than 1080p) runs 5-6fps with Radius 0 and 4fps with Radius 1 (higher quality). Both at once run 4fps (Radius 1) and 5fps (Radius 0). Everything will run faster on a desktop computer with more powerful CPUs+GPU(s).

Thus it's possible to do a complete basic edit and finish without going into another tool and without having to spend time processing DNGs into ProRes/DNxHD. The full version of Resolve is $1000 and its noise reduction apparently isn't as good as Neat Video (from reviews), so doing NR for $200 (or $99 if exporting and processing in PPro/AE/FCP etc.) makes sense.

Resolve has some issues with audio rendering as silence. I was using audio from the internal SSD and video from USB3 external. When I imported audio from same location as video, it began working again.

49s of footage as DNGs and WAV file were 4.62GB on disk, so 4.62GB/49s*60s/1min = 5.66GB/min, *60min/1hr = 339GB/hr, ~2.95hrs on a 1TB drive for ~$80. 4TB drives are around $150. Thus for smaller projects, working directly with DNGs is affordable (if for important work- factor in at least one backup for all the footage).

"Beauty in Nature" was edited from 1TB of material down to 3 minutes. He used a Resolve proxy workflow (edited with proxies in FCP) then relinked to DNGs for final ProRes 4444 (12-bit?) render. He archived all footage to ProRes 4444 (after WB and brightness adjust) and deleted the originals.

In summary: batch-converted DNGs + low-cost hard drives + Resolve 10 Lite + Neat Video + FilmConvert = a pretty decent GPU-accelerated workflow for 5D3 RAW.
 
Simplified Resolve workflow:

1. Download and install 3D LUTs for 5D3 RAW: EOSHD, Hunter, OSIRIS, CineSpace, etc.
2. In Resolve, use the BMD Film Color Space (log-like but not log).
3. In the first node apply the 3D LUT. EOSHD and Hunter get close but are not perfect.

That will get you something decent quickly. Resolve is a lot more powerful than ACR, ACR is a lot easier (initial look).

A calibrated 3D LUT for 5D3 RAW would be helpful (to Rec709 and/or Log). Perhaps Canon will make one? :happy:
 
You can use ACES with the DNG IDT. This will get you standardized image to play with. Not that I see a need for that, other than mixing with footage from other sources.
 
cpc- it would be cool to have a 5D3 RAW to Rec.709 3D LUT. Once in a calibrated color space, it's possible to apply LUTs from REC709 to known film looks, etc. With ACR, it's relatively easy to get realistic/somewhat-accurate looks from 5D3 RAW quickly. It's more challenging to do this with Resolve (starting with the BMD Film color space).

Reviewing test footage, shots in daylight look pretty good in Resolve using the Rec.709 color space. When testing BMD Film color space then applying either the EOSHD or Hunters 3D LUT (both designed for 5D3 footage and BMD Film)- the result is no longer accurate to what was shot (and skintones, for example didn't look as good).

Testing some of the Rec.709 to Log 3D LUTs with Log to Film 3D LUTs works OK sometimes. The Log to Film LUTs with BMD Film haven't worked very well so far (BMD Film is flat but not really Log).

Best results so far using Resolve 10.1.2lite.017:

1. Rec.709 Color Space
2. Custom white balance set to conditions when shot
3. Set Tint to 0
4. Sharpness around 30
5. Highlight Recovery enabled
6. Adjust Exposure as needed

This gets the footage looking as accurate as possible. Using Rec.709 to Film Look/Emulsion 3D LUTs works reasonably well. Rec.709 to Log followed by Log to Film Looks sometimes works OK.

Since the 5D3 RAW footage is close to Rec.709, a calibrated 5D3 RAW to Rec.709 using Resolve's Rec.709 color space would be very helpful (I'd pay for it). I prefer to start with accurate colors then stylize from there (vs. super flat where we can't tell what the correct/original footage looked like. Log makes sense at the time of capture, however 5D3 footage is not Log, so it doesn't makes sense to convert to Log if it can be avoided).
 
You must log in or register to reply here.
Back
Top