HurlBlog Technology Guru: Mike McCarthy Part II
- April 24, 2010
- Shane Hurlbut, ASC
Here is the much anticipated post-production work flow blog that you have been asking for from Mike McCarthy, our technology guru. Please visit Mike’s site at hd4pc.com for even more in-depth technical information on the post-production work flow process.
Post Work flows for DSLRs
“Here at Bandito Brothers, we have been handling the post aspect of many of Shane’s DSLR based projects, ever since the first “Terminiator Webisodes.” The tools available have developed during the past year from a relative hack job, to a reasonably well supported work flow.
The first thing we need to understand about a work flow, is what we are starting with. In the case of Canon DSLR footage, we have full raster HD footage, in YUV 4:2:0, with a full range (0-255) of 8bit color values, at a variety of frame rates. This is saved into Quicktime files, encoded with H.264 compression at about 40Mb/s, with 44.1khz audio. While high bit rate H.264 files preserve a tremendous amount of detail into a relatively small file size, that level of compression makes it difficult to playback the native files in any editing program. In almost all cases it will be easier and more efficient to convert the footage into an intermediate editing format before editorial. This choice of formats will probably be dictated by your NLE options. DNxHD will be the format of choice for Avid, with ProRes for FCP, and a couple other options like Motion-JPEG, MPEG2-IF, or Cineform for Premiere Pro.
From a post perspective, the most obvious unique work flow challenge presented by the original Canon 5DMk2, was “30P!?” Since a transcode to an intermediate format was already required by most work flows, we slowed the footage and the audio by .1% to 29.97 for our first few projects. So 29.97 based work flows can be relatively simple, and are even easier now with the true 29.97 support in the 7D and 1D, and recently the 5D as well.
Inter cutting with film on the other hand usually requires editing and finishing in 24p, (by which I always mean 23.976p) which is a lot more complicated challenge with 5D footage. As Shane has mentioned in the past, the simplest way I have found to deal with this requires that you edit in Avid, and online with Twixtor in AE and Premiere Pro CS4. We use Re:Vision Effects’ Twixtor plug-in to convert our 30p clips to 24p, with true motion compensated frame blending. It works quite well for more footage, but it is extremely render intensive and take a long time to process footage. The details of the relinking process for Twixtored footage with Avid edits are fairly complicated, but can be found on my site, (Link to Avid page on my site) for anyone who is interested in going down that path. For footage shot at 24p on a DSLR, the on-lining process should be relatively straightforward by comparison, and have no unique challenges over 29.97p DSLR work flows.
While Premiere and FCP are both useful tools that will work well on smaller DSLR based projects, Avid is the most stable and responsive editing program, for large projects that encompass hundreds of hours of footage spread across thousands of individual clips. Most Avid edits of DSLR footage will use DNxHD as their editing codec. Since Canon MOVs have a full 0-255 color range, you have to select the RGB (0-255) color space when importing the files into Avid, in order to maintain the full range of the color space. If you are going to use you Avid output as your master, without a separate online conform, using a 10bit editing codec like DNxHD175x will prevent you from losing bit depth during the Rec709 conversion on the initial import transcode. We use 8bit DNxHD36 offline files in our Avid edits, since this is an offline, because we aren’t editing at the 5D’s native frame rate, and we use simple EDLs to online in CS4 via file name relinking after the frame rate conversion. There are other more expensive options for on lining Avid edits, but I am not as familiar with any of them, since Adobe’s Creative Suite satisfies most of our current needs.
Now as a PC guy, I will still be the first to admit that Macs do have their uses. (Specifically generating Pro Res files and accessing HPFS volumes;) For Final Cut Pro work flows, life is a little simpler in that Pro Res is capable of 10bit color by default, as long as the host application supports it. Batching your DSLR files to Pro Res in Compressor should allow you to maintain the full resolution and color space. Compressor also has the capability to solve the 30p to 24p issue through use of Apple’s Optical Flow technology. Compared to Twixtor, our tests have found this process to be slower and the results aren’t quite as good, but if you can’t afford a dedicated conversion plug-in, this is probably the next best thing.
For Premiere based edits, while DSLR files can be played directly on the time line, using an intermediate format will give you a more responsive and stable editing experience. Adobe Media Encoder will give you the proper processing bit depth to convert your files into a variety of possible third party formats, for editing or on-lining in CS4. At Bandito Brothers, we batch process our Canon 5D footage in After Effects, which allows us to use Twixtor to convert our 30p clips to 24p. If the footage is already in the right frame rate, AME is totally sufficient and will process the conversions much faster. We usually online with Cineform AVI files, to utilize the head room that 10bit color offers, especially since SpeedgradeXR can access native files, which is usually our next step after the conform.
Once you have exported an online conform, preferably in 10bit color, there is one more step that should be added to DSLR work flows. There are a number of cleanup processes that can be undertaken to deal with common imaging issues with DSLRs, similar to a dust-busting pass on film work flows. Dead pixels, usually caused by dust on the sensor, which can happen to any camera, are more frequent on DSLRs due to their large sensors and interchangeable lenses. These artifacts are usually static and can be fixed by overlaying nearby pixels that were unaffected, usually directly above or below. You also may see rolling shutter issues, caused by the top of the frame capturing a slightly different moment in time than the bottom. Certain types of rolling shutter artifacts, especially ones related to camera motion, can be fixed with plug-ins from companies like the foundry. Other rolling shutter artifacts like horizontal bands caused by flashes of light are much harder to fix, unless you manually replace the image data with information from another nearby frame. And if you have run a frame conversion process like Twixtor on your footage, this is when you should replace any frames that interpreted poorly with frames from the original source files. These processes are all very labor intensive and require quite a bit of fiddling and fine tuning to perfect your image. As with any step in the process, consider your available resources and carefully prioritize the issues you want to fix.
Once you are finished fixing any defects in the footage, the resulting files should be similar to any other workflow, and you can proceed to visual effects, color correction, tape lay back, web encoding, or disc authoring the same as you would a project from any other acquisition source. Most of the things that are key to an efficient DSLR based work flow take place at the beginning of the process. Once you are off to a proper start, the later steps should come together the same way as any other tapeless project. Hopefully the tips above will give you a good overview of the potential pitfalls, and things you can do to stay one step ahead of the game. I have much more detailed information available on my site www.hd4pc.com/techblog/2010/04/24/editing-dslr-footage-in-avid/, that I will continue to update as new developments are released.”