ProResSD - On a G5? You might want to avoid it

After the results of the ProResSD tests I performed, I've started using it in situations where pristine quality wasn't a concern. And the more I use it the less likely it is I'll continue to use it.

This has nothing to do with its quality and everything to do with the fact that I'm still on a G5 (Dual 2.5).

If I do *anything* to ProResSD compressed images, the image quality drops to "Preview" - meaning I have to render before doing any outputs. After 18 months of working on this machine in which I can often have two 3-Way Color Correction filters, plus Broadcast Safe, with a crop or reposition and have everything playback at full quality with no rendering - having to force a render for even the slightest repo is driving me nuts.

I can safely say I'll be using ProResSD a lot less than I thought I would. Considering that a Quad-Core is in the near-term future I don't have the time or inclination to deal with the extra overhead forced upon me by this new codec.

If anyone has any experience using ProResSD on a Mactel, please drop a message in the Comments box. I'd love to know your experiences.

- pi

 Subscribe in a reader

|

ProRes SD Results (Finally)

2 weeks ago I finished my testing on the ProRes SD flavor of Apple's new codec. While many have tested the ProResHD variant, the SD variant hasn't been quite as scrutinized. Perhaps that's because hard drives have gotten so large and pipelines so wide that a 25 mb/s files isn't the bottleneck it used to be? Still, if you can save the space without giving up quality why not jump off the Uncompressed codec and onto the ProRes bandwagon when working at Standard Definition? To make an informed jump, we need some facts. What follows is the result of some of my "fact-checking" as I determine if ProRes SD is worthy as a 'finishing' codec and can withstand a common finishing workflow.

And if you want more info on the ProRes codec, here's the direct download of
Apple's ProRes white paper.

Judging Criteria: To judge if ProResSD was a "finishing" codec I decided I had to be able to cut, mid-shot, the original 10bit textless back into the 3rd Generation ProResSD Protection Master - as if I were creating an International Generic Master. And at the edit point it had to have no visible difference to both the human eye and the waveform/vectorscope. This is a test I know a fully 10bit uncompressed workflow could easily pass. And frankly, this is not a very challenging test even for an analog tape format like D-2 (assuming an all-digital environment). So my judging on ProResSD will be fairly harsh - it needs to be perfect.

Methodology:
Using a reality series I finished earlier this year as reference footage - I created a 2 minute test sequence comprised of interiors, exteriors, day, night, interview and run & gun situations. The footage was originally shot anamorphically on DVCPro, conformed in an Avid at 1:1 and then it was output to Digibeta for final finishing in our FCP finishing bay. I captured the footage via Decklink HD Pro SDI to 4 codecs:
  • 10bit Uncompressed
  • ProRes SD (High Quality)
  • ProRes SD (Standard Quality)
  • DV
The footage was then color corrected using Apple's new Color app, and rendered back its corresponding codec. The exception here was DV, which I "promoted" to a ProRes HQ sequence before sending it to Color and then treated as an HQ sequence from there on. I was curious if ProRes would further degrade the DV material with its inherit macroblocking issues. I color corrected the 10bit sequence first, for subsequent sequences I imported the corrections from the 10bit pass.

Simulating the worst-case scenario for a show being delivered to a network - I assumed the footage would be output and recaptured several times:
  • Textless Output
  • Textless Captured, Master Output
  • Master Captured, Protection Output
  • Protection Captured, International Generic created
Finally, I used the 10bit uncompressed textless as the base comparison, differencing it with the 'International Generic' of each of the other three codecs to identify the most challenging / degraded images.

Difference Tests (images will open in new windows):

  • Digibeta Capture: download image (2MB)
    I did this series of difference tests mostly from curiosity. It compares the 10bit Uncompressed to each of the other three codecs before any other processing. It gives an idea as to how much detail each codec throws away. If you've ever wondered why so many of us despise working with DV, every bit of detail you see in these tests is detail retained by the 10bit Uncompressed codec and thrown away by the DV codec.
  • Color Correct Output: download image (2MB)
    After rendering the color correction out from Color, I did another series of differences versus the 10bit render. I was looking for any obviously increased degradation that wasn't seen in the first set of Digibeta Capture difference tests. I don't see any. Color seemed to render them cleanly - especially the DV rendered out as ProResHQ which didn't seem to suffer any additional degradation, leaving open one interesting workflow possibility for those with constrained budgets and originating from DV.
  • 3rd Generation Tests: download image (2MB)
    After round-tripping from FCP to Digibeta 3 times, I again made a series of differences, this time between each codecs Textless and its 3rd generation - seeing how well it held up. The 10bit Uncompressed was rock-solid black, so I didn't bother to include it here.

Frankly, I was surprised how well the DV Promote workflow held up. After the initial hit during capture, the ProResSD didn't allow it to degrade any further. In my opinion, this is a viable workflow for DIY'ers who don't have SDI workflows available to them. But as you can probably see from the difference tests, the ProResSD is indeed lossy - but we already knew that, Apple doesn't make any claims otherwise. Which brings me back to where I started:

Conclusion

Q: Can a 3rd generation copy be visually distinguished when edited mid-shot into a 1st generation copy or can it be easily observed using a waveform monitor or vectorscope?

A
: The answer to both parts of that compound question is... When playing at speed, 1st Generation 10bit is indistinguishable from 3rd Generation ProResSD. I can't see the edit. By that standard ProResSD is indeed a finishing codec, even as we know there's been slight generational loss as observed in the difference tests.

But: When paused on identical frames and quickly toggling between 1st generation Uncompressed and the 3rd generation ProResSD - levels and chroma are rock solid steady, but there is a oh-so-slight softening of the image. It's slight enough that most my clients won't be able to see it. Heck, I barely see it. Though once I noticed it on the monitor and l looked back at my scopes, I could see a teeny softening of the trace. It wasn't evident in every shot, only those with heavy details (usually in the background). So...

ProsRes SD is an impressive codec. While only doubling the storage space of DV it gives 98% of the quality of Uncompressed. Good enough for finishing purposes? Yes. I would not use it for heavy compositing where every drop of detail is essential. Unlike the HD variant, which I've heard is rock-solid through (at least) 10 generations, the SD variant's 'lossy-ness' does exist after 3 generations.

And here's where the rubber meets the road: Will I be using it as my codec of choice? Not for network deliverables. I want my images as pristine as possible and with storage space so cheap, 25 MB/s isn't that big a deal anymore. But I
will use it for creating DVD, web deliverables, screening copies, etc - replacing 8bit uncompressed as my codec of choice for those elements. And on low budget projects without compositing needs - I'm sure there will be a few projects where I will advocate capturing ProResSD to use it from the first Assembly through the end to final Master.

UPDATE: If you're running on a G5, be sure to read
this follow-up post why Pro-Res isn't quite so thrilling on those machines.

- pi

 Subscribe in a reader

|

ProRes SD Update

In case anyone is playing along... I was about to re-do my portions of my workflow test when the dot upgrade to FCS2 rolled out. I put it on hold and have been busy with clients since then.

I have re-worked my testing workflow to ensure my results are reliable - but the holiday is upon us and I'm off the rest of the week. Next Monday or Tuesday I should be back on this side-project.

- pi

 Subscribe in a reader

|

ProRes SD in Practice

UPDATE: Testing has been completed. Different conclusions have been drawn. Final results are here.

One of the new workflows introduced by Apple in Final Cut Studio 2 is a lightweight codec called ProRes 422. According to Apple's ProRes White Paper:

Apple ProRes 422 is changing the rules of post-production. The combination of industry-leading image quality, low data rates, and the real-time performance of Final Cut Studio 2 makes ProRes 422 the ideal format to meet the challenges of today’s demanding HD production workflows.

If you read the White Paper the emphasis is almost entirely on HD, even though an SD variant ships with FCS2. After some testing I think I understand why...

ProRes422 SD seems quite lossy. After 3 generations I'm seeing a definite softening of details. It's graceful, similar to analog degradation in the more modern analog tape formats, but it's there. It's enough loss to say I don't consider it a finishing codec - I'll be staying uncompressed. For editors out there who ran digital analog component rooms - I'd compare it to D2 running through a digital switcher. I used to go 6-8 generations on that format in a well-designed edit bay. I didn't take ProRes SD that far - but I don't have much hope it would fare any better.

I'll be posting a full-blown review of ProResSD in the next two weeks - but one word of warning about a proposed workflow I've seen discussed online: Putting DV footage into a ProRes timeline (or "promoting" as you would into an Uncompressed 8-bit or 10-bit timeline) is a good way to give your footage an untimely death. I'll be re-testing my results to be sure, but for now I'd advise against it (especially if you plan on running it through Color).

 Subscribe in a reader

|