Superproducer Ted Hope on...

June 16, 2010

Is 5DtoRGB the DSLR post-production solution we've been waiting for?

As part of today's free Adobe workshop on HDSLR feature film workflows (I'll update that post with a link to the archived presentation), one of the commenters brought to my attention a promising DSLR post-production plugin called 5DtoRGB. ((Sorry, whoever it was who commented about 5DtoRGB -- I lost the chat thread when my Flash plugin crashed for the dozenth time.)) I've called DSLR color-correction on a Mac a clustercuss, not just because of the 4:2:0 color space, but because of Quicktime gamma inconsistencies (often related to YUV to RGB conversion). Rarevision's 5DtoRGB plugin, currently in beta, just might be the solution to these problems.

Here's the word from Rarevision:

Many programs use QuickTime internally to perform YUV to RGB conversion which, according to our testing, does only a mediocre job... You're stuck with it if you've converted your footage with Final Cut Pro, Adobe After Effects, MPEG Streamclip or Canon's Movie Plugin-E1 for Final Cut Pro -- all of them use QuickTime to decompress H.264. 5DtoRGB takes a no-compromise approach to quality, ignoring any concerns about speed. 5DtoRGB bypasses QuickTime decoding altogether, works internally at 10 bits and uses double-precision floating point math for its YUV to RGB conversion. It also recognizes Canon's full range 8 bit YUV values (0-255), avoiding clipping and the resulting loss of picture information. The resulting files are the absolute highest quality you'll ever get out of the camera. In fact, you could argue that they're even better than the camera originals since they've undergone high quality chroma smoothing (which you can disable if you want, but you shouldn't).

Presumably this plugin will be much slower than Quicktime-based plugins, and would be overkill for quick jobs or content destined for the web. But for anyone interested in shooting a feature film -- or anything requiring heavy post work or color-correction -- 5DtoRGB just might be the ticket. I've downloaded the beta plugin but can't currently test it as I'm moving (and don't currently have access to my camera or hard drives) -- woe is me. Beyond that, the plugin requires a 64-bit processor, which I also don't have (yet). However, take a look at these blown-up frame grabs -- the first is from Canon's very convenient E1 EOS plugin (which presumably produces the same results as Magic Bullet Grinder, given they're both Quicktime-reliant), and the second is from 5DtoRGB:

Projected in a theater, or viewed on a good HDTV, this difference will definitely be noticeable. Not to mention the greater flexibility a 4:4:4 file will give you in post. There are all sorts of questions that spring to mind related to 5DtoRGB -- transcode time, plugin cost, storage needs, and what a sensible workflow would be (edit native h.264 files or transcoded ProRes files, treating them like proxy files, eventually swapping in the 5DtoRGB files?), but I'm exciting about the possibilities of squeezing the best possible quality out of the 5D for use on my forthcoming feature-length project.

If anyone's had a chance to work with the beta plugin, please share!

5DtoRGB by Rarevision »

Your Comment

30 Comments

Haven't used it, but like you, I don't have a 64-bit proc.

June 17, 2010

0
Reply
jamison

Do you know if that session yesterday was recorded?

June 17, 2010

0
Reply
Stephen Henderson

Haven't used it, but those samples look amazing.

Any idea on pricing when the final product will come out?

June 17, 2010

0
Reply
Alam

No official word on pricing, but they did say via email "The pricing will be very reasonable. It'll be well within everyone's price range. :-)"

Stephen, try the Adobe post (linked above) for a link to the archive.

June 18, 2010

-1
Reply
avatar
Ryan Koo
Founder
Writer/Director

Ryan, thanks for the followup post--always appreciate work towards solutions...

But I am still bloody confused about the ideal DSLR post workflow. Being the obsessive perfectionist sort, and being relatively new to the NLE/DSLR game--makes for frustration.

Two things I'm not clear on:

1. I've been told from very knowledgeable Codec People that there is little to gain using 4:4:4 for DSLR 4:2:0 footage. 4:2:2 is another story, they say. Poppycock? Either way, to me, at least for the initial cut, 4:2:2 seems a wiser choice.

2. 5D to RGB apparently offers 422 outputs as well as 444 (like ProRes LT or other variants) which are YUV--right? Would YUV Prores files generated through these also superior to the EOS plugin outputs, by using a quicktime-independent conversion? Or does that superiority only apply to true 444 RGB files?

My nagging productive conscience says to forget about this whole discussion, get to editing, and once everyone has it figured out, swap out my Prores LT files for whatever superior RT solution is out there when I'm ready to do final color correction. The perfectionist in me wants an answer now!

What about Cineform's Neo HD--is that a good solution to this mess you all speak of? I know they avoid using Quicktime to create the Cineform master files, which are fully RT in FCP. But what about the Prores output options through Neo HD--does Cineform, like all the rest, also use Quicktime to create the Prores files and therefore clip some highlights or miss on some colors as you describe?

Do Snow Leopard and FCP 7 resolve most of these issues, at least within an editor's own world of post-production, before others view the footage on their own weird monitors?

Like I said, I'm about to move on with a solution for now, but would really appreciate hard and fast answers to any of these questions as soon as possible.

Thanks for all of your sharing of knowledge here, Ryan, much obliged. All best, JB

June 18, 2010

0
Reply

Joe, that's why I've referred to it as a cluster***!

I think your instincts here are right on: "get to editing, and once everyone has it figured out, swap out my Prores LT files for whatever superior RT solution is out there when I’m ready to do final color correction."

Because: no, Snow Leopard and FCP 7 do NOT resolve most of these issues in my experience, and Cineform won't be out for CS5 until next month. Neo HD is $500 too, so it might be worth trying Neoscene instead, which is $129. I wish there were "hard and fast" answers but if my own experiences (as well as others') have proven anything to date, it's that there is no such thing.

As a final note, I do think 4:2:2 is the best compromise. The issues relating to gamma inconsistencies are independent of that, I believe (since they're a Quicktime issue, regardless of what color space you end up in).

June 18, 2010

0
Reply
avatar
Ryan Koo
Founder
Writer/Director

Thanks, Ryan.

That YUV vs. RGB question may be one of semantics, but isn't it supposed to explain why the "5D to RGB" framegrab looks better than the EOS plugin generated one, according to Rarevision? I'm still just as confused because ProRes isn't RGB, it's YUV...right? I think I've just tripped down the psychedelic rabbit hole...

I do understand your point about the gamma inconsistencies being a Quicktime issue--everyone seems to agree on that. We probably ought to let it lie, or switch platforms if we're that annoyed by it.

But what about the subject of this post, namely, the overall quality of Prores (YUV) generated through different programs from DSLR source footage? Could someone more clearly explain the importance of colorspace (YUV/RGB) with regards to generating ProRes files?

My simple, neophyte imagination would think that the Gamma inconsistency is to blame on that cluster of issues as well--the Quicktime program doesn't recognize a certain range, and clips those colors, resulting in an inferior image on the end conversion. This is probably a simplistic, completely wrong interpretation, but it's the best I can manage not having taken courses on Color Space and codecs and being new to the post game.

If Neoscene or Neo HD create Prores without Quicktime in the conversion, then we may already have a solution. Alternatively, we could just go with their Cineform 422 codec instead of ProRes--which I'm apt to do except I'm working with a Mac and a Matrox MXO2, and I'm not sure that combination of hardware supports First Light, which is Cineform's very cool nondestructive metadata color correction tool. It may still be worth making the switch to Cineform anyway, though. "If you value your image" as they say...

If whomever is reading this isn't getting impatient, I am, so I'll stop there. I'm giving myself thirty minutes to decide, and I'm back to editing. Swap out later if I change my mind. But jeez this is annoying.

Thanks, cheers.

June 18, 2010

0
Reply

I've just been told by a knowledgeable friend that all third parties must use Quicktime to generate ProRes files...has Rarevision bent the rules here somehow? Or is it the smoothing function or compression that makes the file look better.

Most everything I've read seems to point to Cineform's wavelet-based compression being the best. And now it has RT support in FCP. If I could just get First Light monitoring through Matrox to HP Dreamcolor I would be a happy dude.

I'm off to run some quick tests, my thirty minutes is almost up. Thanks for opening this annoying can of worms, Ryan. No thanks for myself for overresearching just about everything I do. Gracias, JB

June 18, 2010

0
Reply

Sorry it's such a PITA, Joe. I wish I had a good solution, but I don't. I've emailed both Cineform and Rarevision asking for clarifications, but I'm not sure if we'll get a response -- since I switched servers, most of my emails seem to be going to people's spam folders. Perhaps, like you, I can overresearch this (spam) problem only to find there's no easy solution!

June 18, 2010

0
Reply
avatar
Ryan Koo
Founder
Writer/Director

Ryan,

I'm giving Cineform a shot, as it's a solution that's available now, and isn't going anywhere. Before I stick with it for everything, I'm going to try to run some comparison sequences against ProRes LT.

I really like Firstlight so far--I can Firstlight CC on my Mac Display, then double check through the Matrox/Dreamcolor, as all First Light color changes are made immediately apparent in other applications like FCP's use of a particular video file. Obviously I'd prefer to monitor simultaneously within FirstLight on my broadcast monitor in addition to my desktop one--hopefully ultimately there will be a way for me to do that. But in the meantime this could be viable, allowing me to make some good use out of the full Cineform suite if I end up saving up for the whole (permanent) package.

Cineform's files sizes are bigger than ProRes LT (a drawback). But theoretically they may not suffer from the weird compression inconsistencies referenced in this post that pop up in QT based conversions for DSLR. They will probably, however, still be subject to QT player gamma shifts--all codecs are, from what I understand.

Neo HD comes at a cost, but First Light seems very intuitive, from what I can tell so far from the trial edition: a lot of fun to play with, and less scary to me than rendering CC. Perhaps, if you're experienced, FCP's 3 way CC and filters work just as well, but as someone new to post, First Light seems like a nice, fresh approach to quickly get the look you're after without committing too much to that look by rendering or burning in filters.

But, who knows; it seems an online search through the forums can get a vote for every conceivable point of view ("Streamclip is better/streamclip is better and faster" "E1 plugin is better, while slower." "They're all the same." Same with the "Prores is better" "Cineform is superior" back and forth). A wiser man, as I noted above, would have moved on to editing by now and worried about perfection later. But I guess it's not really about logic, is it--you shoot this stuff, you want it to look as good as possible right NOW, not decent to be improved at a later date...

And for that there's a great Bob Dylan lyric: "You can't open up your mind to every conceivable point of view" I believe it goes. A online search might corroborate--or disprove--that. But it seems a personal test is really the only way to go. Luckily for us there are trial editions.

All of that said, I would very much appreciate someone far more experienced than I writing a detailed analysis of these issues, with a great concise summary at the end for everyone to work from. Not that it wouldn't inevitably contain some personal bias toward one workflow or the other...anyone want to throw in for an independent panel of colorists to candidly assess the options available to the DSLR indie?

Thanks again, and apologies for my longwindedness. Appreciate your blog as a friendly, organized forum for sorting through post production demons; hopefully someone else stands to benefit, even just a little from these my demented addendums to yours. Peace, JB

June 18, 2010

0
Reply

5DtoRGB uses QuickTime to COMPRESS to ProRes, not DECOMPRESS from H.264. There's no way around this -- you have to invoke Apple's APIs to get access to the ProRes codecs. Until ProRes is reverse-engineered, this will remain the case, as Apple wants it.

The reason 4:2:2 files look better from 5DtoRGB is because the H.264 decompression step is superior to QuickTime. We don't use QuickTime to decompress H.264. QuickTime's H.264 decompression is awful, and this is true no matter which codec you transcode to -- as long as H.264 material is decoded through QuickTime, the fault is there and it's permanent.

5DtoRGB was originally designed for DPX output for film and VFX work (hence the name, "5DtoRGB"). We added ProRes output because everyone wants it. We mostly do finishing work for indie productions, so we have a lot of experience working with cheap cameras and consumer/prosumer camera formats. Our job is to make this stuff look as good as possible, hence our need to develop proprietary tools. With the popularity of the Canon SLRs for video, I decided others could benefit from this research. Our main gig is post, not software development -- so, 5DtoRGB will be very affordable as well as every other tool we release (dMatrix is free, for example).

On the 4:2:2 vs 4:4:4 thing -- In a perfect world, we will all be operating in RGB. All displays now are RGB displays. Compositing apps like After Effects work in RGB. 4:2:2 compression is no longer necessary since we have much better technologies now for compressing picture information. This subsampled chroma thing is a bad leftover from the early days of color broadcast, and hopefully will go away someday. We try to work in 4:4:4 as much as possible. ProRes 4:4:4:4 is a nice addition to the ProRes line, especially with its alpha support. The problem with 4:2:2 is it adds a bunch of processing steps to recover RGB, all of which can vary based on software implementations, matrices, etc. If there were no 4:2:2 video, you would not have all these weird gamma problems with QuickTime. 4:2:2 is antiquated, but we're stuck with it because of grandfathered broadcast standards.

June 18, 2010

0
Reply

Thanks for the comments Joe and Thomas. For those following along who might want a summary of what's transpired:

-- Quicktime has problems with h.264 footage, no matter what transcoding software you use (Canon E1, Magic Bullet Grinder, MPEG Streamclip). There's no quick fix for the problem.

-- 5DtoRGB's advantage is that, unlike the aforementioned plugins, it avoids Quicktime to DECOMPRESS h.264 (and then converts to ProRes/DPX). Thomas is saying Quicktime's h.264 decompressor is shoddy, which is why no matter which format you RECOMPRESS to, if the first step is reliant on QT, the recompressed files won't be of the best quality.

I look forward to getting my hands on 5DtoRGB when it's released -- hopefully by then I'll have a computer that can handle it.

June 18, 2010

1
Reply
avatar
Ryan Koo
Founder
Writer/Director

I misspoke about the size of Cineform 422 (High). In my tests so far they are a bit smaller than equivalent Prores LT files. I'm hoping to write a brief article about my sorting through these options; so far I'm pretty impressed with Cineform and Firstlight. On the Prores side, as great as MPEG Streamclip is for a variety of needs, it seems to shift the colors for DSLR sources in a substantially more pronounced way than Apple Compressor in FCP 7, whose LT files look fairly accurate to the original. The Cineform ones add a little more range it seems, especially in the blacks, while allowing rapid CC in firstlight. I'd be curious about 5D to RGB as well, once they have aomethig that can convert full length files. Obviously I broke my promise to myself to quit fiddling with this...Thanks for your posts Ryan, and follow up, Thomas. Cheers, JB

June 23, 2010

0
Reply

No words, just to pics.

Original H.264 from Canon: rozhabrev.ru/test-canon.png
5DtoRGB 4444: rozhabrev.ru/test-444.png

June 23, 2010

0
Reply

Very interesting Ro! Thanks for posting. From what I can tell:

--Canon has a lot more chroma noise in the red channel
--Canon appears to have a bit more noise in general (looking to the right of the cup)
--5D2RGB, at first glance, does less chroma smoothing than Canon -- the golden arches, for example, appear more jaggedy in the 444 version -- but I wonder if that's because its actually retaining more sharpness

By "Canon," of course, I really mean "Quicktime" since it's QT's h.264 decompression that is the culprit. Very interesting to see -- thanks very much for sharing!

June 23, 2010

-1
Reply
avatar
Ryan Koo
Founder
Writer/Director

You are welcome! I have some new picts.

Compressor's Prores 422 with noise reduction: rozhabrev.ru/test-compressor422.png
Canon EOS FC plug-in's prores 422: rozhabrev.ru/test-plugin422.png
Compressor's Prores 4444 with noise reduction: rozhabrev.ru/test-compressor4444.png
Canon EOS FC plug-in's prores 4444: rozhabrev.ru/test-plugin4444.png

Seems like except gamma problem, compressor gives best quality, or nearly the same as 5D2RGB. Am I wrong?

June 28, 2010

0
Reply

Very interesting, Ro. I'll see if I can get a comment from Rarevision. For the benefit of others, here are your links in clickable form (keep in mind this is being viewed at 800% -- the 5D's compression isn't nearly as noticeable at 100%):

Compressor vs 5DtoRGB in 4:2:2:
http://rozhabrev.ru/test-compressor422.png
http://rozhabrev.ru/test-plugin422.png

Compressor vs 5DtoRGB in 4:4:4:
http://rozhabrev.ru/test-compressor4444.png
http://rozhabrev.ru/test-plugin4444.png

What kind of noise reduction were you using?

June 28, 2010

0
Reply
avatar
Ryan Koo
Founder
Writer/Director

Comment with links wasn't added. I can't use links in comments, right? Ok. Again.

Ryan, links in your repost aren't right named. "Plugin" in file name means E1 EOS plugin, not the 5DtoRGB. And I missed 5DtoRGB 422 prores. So here are all right links:

Original H.264
http://rozhabrev.ru/test-canon.png
E1 EOS plugin
http://rozhabrev.ru/test-plugin422.png
http://rozhabrev.ru/test-plugin4444.png
Compressor
http://rozhabrev.ru/test-compressor422.png
http://rozhabrev.ru/test-compressor4444.png
5DtoRGB
http://rozhabrev.ru/test-422.png
http://rozhabrev.ru/test-444.png

June 29, 2010

0
Reply

Ro, If you just add "http://" to these links they should be correctly interpreted. I've done it for ya -- thanks again for the caps.

June 29, 2010

0
Reply
avatar
Ryan Koo
Founder
Writer/Director

Ro, thanks for sending me your clip. You should be getting results such as these:

http://rarevision.com/5dtorgb/mcd_neo.png
http://rarevision.com/5dtorgb/mcd_5dtorgb.png

The first is your clip converted with CineForm's NeoScene. The second is converted with 5DtoRGB. Both were converted to ProRes 422 QuickTime files. ProRes 4444 looks slightly better, since it is not subsampled. We recommend using ProRes 4444 whenever possible.

These were viewed with Adobe After Effects. Final Cut Pro sometimes does not display footage with good chroma interpolation in its viewer.

July 2, 2010

0
Reply

That looks great Thomas. Looking forward to 5DtoRGB's availability.

July 6, 2010

0
Reply
avatar
Ryan Koo
Founder
Writer/Director

>We've got a new beta up at http://rarevision.com/5dtorgb/. There is no processing limit now, nor any time limit.

August 9, 2010

0
Reply

I just downloaded and tested the plug-in today. It works VERY well. Definitely slower, but I ran some particularly difficult camera originals through it from a recent project of mine. The results were noticeably smoother and cleaner with regard to banding and artifacting compared to other options I have tried (Mpeg Streamclip, Compressor, Grinder). The color gamma of the resulting prores files also seems to remain consistent from application to application (Nuke, FCP, After Effects, Shake) not to mention both old and new versions of Quicktime.

August 29, 2010

0
Reply

It really really does. I've done a proper test and writeup with images and stuff on the blog and I was really really impressed with it.

http://www.elskid.com/blog/?p=1647

August 30, 2010

0
Reply

Great stuff Robin

August 30, 2010

0
Reply
avatar
Ryan Koo
Founder
Writer/Director

Am I to assume that, since the title of the plugin is called 5DtoRGB, support for the 550D isn't there? Robin's comparisons make me a believer.

August 30, 2010

0
Reply
Nick

The name's the worst thing about the software! H264 files are H264 files and that's it, so yes it'll work on any camera in the range, from 550D to 1D, all of them shooting to the same file format.

August 30, 2010

0
Reply

Just done a massive follow up test on that 5DtoRGB plugin with footage going to Prores SQ... kind of wish I hadn't. The conclusion I can draw is that, contrary to popular belief, bumping the files up to 4444 through 5DtoRGB actually does make a difference. Depressing experience doing this test, and not what I expected at all!

http://www.elskid.com/blog/?p=1700

Cheers!

R

September 2, 2010

0
Reply

Excellent work, all. I was able to test out the beta earlier in the week with my 550D files, and while I was a little disappointed at not having access to 4:4:4:4 (I'm an FCS 2 guy at the moment), the 4:2:2 compression blew me away. Hands down it was leagues better than anything done with Compressor. I believe its largely due to the chroma smoothing at work, but I could be wrong.

September 2, 2010

0
Reply
Nick

Koo,
Looks like 5DtoRGB wins over MPEG Streamclip, but how do you find that 5DtoRGB stacks up against Magic Bullet’s Grinder?

August 11, 2011

0
Reply
Caleb V.