A case study on dealing with Canon 5D Mk2 footage on actual productions.
You could say that it started with Panasonic and Nikon, but it wasn’t until professional photographer Vincent Laforet posted his ground-breaking short film Reverie, that the idea of a shooting video with a DSLR (digital single lens reflex) camera caught everyone’s imagination. The concept of shooting high definition video with a relatively simple digital still camera was enough for Red Digital Cinema Camera Company to announce the dawn of the DSMC (digital still and motion camera) and push it to retool the concepts for its much anticipated Scarlet.
The Scarlet has yet to be released, but nevertheless, people have been busy shooting various projects with the Canon EOS 5D Mark II like the one used by Laforet. Check out these projects by directors of photography Philip Bloom and Art Adams. To meet the demand, companies like Red Rock Micro and Zacuto have been busy manufacturing a number of accessories designed specifically for the Canon 5D in order to make it a friendlier rig for the operator shooting moving video.
Frame from Reverie
Why use a still camera for video?
The HOW and WHY are pretty simple. Digital camera technology has advanced to the point that full-frame-rate video is possible using the miniaturized circuitry of a digital still photography camera. Nearly all DSLRs provide real-time video feedback to the LCD display on the back of the camera. Canon was able to use this concept to record the “live view” signal as a file to its memory card. The 5Dmk2 uses a large “full frame 35mm” 21.1 MP sensor, which is bigger than the RED One’s sensor or a 35mm motion picture film frame. Raw or JPEG stills captured with the camera are 5616×3744 pixels in a 5:3 aspect ratio (close to HD’s 16:9). The video view used for the live display is a downsampled image from the same sensor, which is recorded as a 1920×1080 high-def file. This is a compressed file (H264 codec) at a data rate of about 40Mbps. 16:9 is slightly wider than 5:3, so the file for the moving image is cropped on the top and bottom compared with a comparable still photo.
The true beauty of the camera is its versatility. A photographer can shoot both still images and motion video with the same camera and at the same settings. When JPEG images are recorded, then the same colorimetry, exposure and balance will be applied to both. Alternatively, one could opt for camera raw stills, in which case the photos can still be adjusted with great latitude after the fact, since this data would not be “baked in” as it is with the video. Stills from the camera use the full resolution of this large sensor, so photographs from the Canon 5D are much better than any stills extracted from an HD camera, including the RED One.
Frame from Reverie
Videographers have long used various film lens adapters to gain the lens selection and shallow depth-of-field advantages enjoyed by film DPs. The Canon 5D gives them the advantage of a wide range of glass that many may already own. The camera creates a relatively small footprint compared to the typical video and film camera – even with added accessories – so it becomes a very interesting option in run-and-gun situations, like documentaries. Last but not least, the camera body (no lenses) costs under $3K. So, compared with a Sony EX3 or a RED One, the 5Dmk2 starts to look even more attractive to low-budget filmmakers.
What you lose in the deal
As always, there are some trade-offs and the Canon EOS 5D Mark II is no exception. The first issue is recording time. The Canon 5D uses CF (CompactFlash) memory cards. These are formatted as FAT32 and have a 4GB file limit. Due to this limit, the maximum clip length for a single file recorded by the 5Dmk2 is about 12 minutes. Unlike P2 or EX, there is no provision for file spanning. The second issue is that the camera records at a true 30fps – not a video friendly 29.97 and not the highly desirable film rate of 23.98 or 24fps.
Audio is considered passable, but for serious projects, double-system, film-style sound is recommended. This workflow would be the same as if you were shooting on film. Traditional slates and/or software like PluralEyes (Singular Software) or FCPauxTC Reader (VideoToolshed) make post syncing picture and sound a lot easier.
Example of the rolling shutter effects used for interesting results
One major limitation cited by many is the rolling shutter that causes the so-called “jello” effect. The Canon 5D uses a single CMOS sensor and nearly all CMOS cameras have the same problem to some degree. This includes the RED One. This image artifact arises because the sensor is not globally exposed at the same point in time, like exposing a frame of 35mm film. Instead, portions of the sensor are sequentially exposed. This means that fast motion of an image or the camera translates into the image appearing to wobble or skew. In the worst case, the object in the frame takes on a certain rubbery quality, hence the name the “jello” effect. It can also show up with strobes and flashes. For example, I’ve seen it on strobe light and gun shot footage from a Sony EX3. In this case, the rolling shutter caused half of the frame to be exposed and the other half to be dark.
Skew or wobble becomes most obvious when there are distinct vertical lines within the frame, such as a lamp post or the edge of some furniture. Fast panning motion of the camera or subject can cause it, but it’s also quite visible in just the normal shakiness of handheld shots. If you notice many of the short films on the web, the camera is almost always stationary, tripod-mounted or moving very slowly. In addition, lens stabilization circuitry can also exacerbate the appearance of these artifacts. Yet, in other instances, it helps reduce the severity.
Note the skew on the passing subway cars
High-end CMOS cameras are engineered in ways that the effect is less noticeable, except in extreme circumstances. On the other hand, the Canon 5D competitor – the Nikon D90 – gained a bit of a reputation specifically for this artifact. To combat this issue, The Foundry recently announced RollingShutter, an After Effects and Nuke plug-in designed to tackle these image distortion problems.
Don’t let this all scare you away, though. Even a camera that is more subject to the phenomenon will turn out great images when the subject is organic in nature and care is taken with the camera movement. Check out some of the blog posts, like those from Stu Maschwitz, about these issues.
Frame from My Room video
But, how do you post it?
Like my RED blog post, I’ve given you a rather long-winded intro, so let’s take a look at a real-life project I recently posted that was shot using the Canon EOS 5D Mark II. Toby Phillips is a renowned international director, director of photography and Steadicam operator with tons of credits on commercials, music videos and feature films. I’ve worked with him on numerous spots where his medium of choice is 35mm film. Toby is also an avid photographer and Canon owner (including a 5D Mark II). We recently had a chance to use his 5Dmk2 for a good cause – a pro bono fundraiser for My Room, an Australian charity that assists the Children’s Cancer Centre at the Royal Children’s Hospital in Melbourne. Toby needed to shoot his scenes with minimal fuss in the ward. This became an ideal situation in which to test the capabilities of the Canon and to see how the concept translated into a finished piece in the real world.
Frame from My Room video
Toby has a definite shooting style. It typically involves keeping the camera in motion and pulling focus to just hit a point that’s optimally in focus at the sweet spot of the camera move. That made this project a good test bed for the Canon 5D in production. Lighting was good and the images had a warm and appealing quality. The footage generally turned out well, but Toby did express to me that shooting in this style – and shooting handheld without any of the Red Rock or Zacuto accessories or a focus puller – was tough to do. Remember that still camera lenses are not mechanically engineered like a motion picture lens. Focus and zoom ranges are meant to be set and left, not smoothly adjusted during the exposure time.
Posting footage from the 5Dmk2 is relatively easy, but you have to take the right steps, depending on what you want to end up with. The movie files recorded by the camera are QuickTime files using the H264 codec, so any Mac or PC QuickTime-compatible application can deal with the files. They are a true 30fps, so you can choose to work natively in 30fps (FCP) or first convert them to 29.97fps (for FCP or Avid). That speed change is minor, so there are no significant sync or pitch issues with the onboard audio. If you opt to edit with Media Composer, simply import the camera movies into a 29.97 project, using the RGB import settings and the result will be standard Avid media files. The camera shoots in progressive scan, so footage converted to 29.97 looks like that shot with any video camera in a 30p mode.
Canon 5D and Final Cut Pro
I edited the My Room project in Final Cut. Although I could have cut these natively (H264 at 30fps), I decided to first convert the files out of H264 for a smoother edit. I received the raw footage on a FireWire drive containing the clips copied from the CF cards. This included 150 motion clips for a total of about one hour of footage (18GB). The finished video would use a mixture of motion footage and moves on stills, so I also received another 152 stills from the 5Dmk2 plus 242 stills from a Canon G10 still camera.
Step one was file conversion to ProRes at 1920×1080. Apple Compressor on a MacBook Pro took under five hours for this step. Going to ProRes increased the storage needs from 18GB to 68GB.
Step two was frame rate conversion. The target audience is in Australia, so we decided to alter the speed to 25fps. This gives all shots a slight slomo quality as if the footage was shot in an overcranked setting. The 5Dmk2 by itself isn’t capable of variable frame rates or off-speed shooting, so any speed changes have to be handled in post. Although a frame rate change is possible in the Compressor setting (step 1), I opted to do it in Cinema Tools using the conform function. When you conform a file in Cinema Tools, you are altering the metadata information of that file. This tells a QuickTime-compatible application to play the file at a specific speed, such as 25fps instead of 30fps. I could also have used this to conform the rate to 29.97 or 23.98. Because only the metadata was changed, the time needed to conform a batch of 150 clips was nearly instantaneous.
Step three – pitch. Changing the frame rate through conform slows the clips, but it also affects the sync sound by making it slower and lowering the pitch. Our video was cut to a music track so that was no big deal; however, we did have one sync dialogue line. I decided to fix just the one line by using Soundtrack Pro. I went back to the original 30fps camera file and used STP’s TimeStretch. This let me adjust the sync speed (approximately 83% of the original) to 25fps, yet maintain the proper pitch.
Step four – stills. I didn’t want to deal with the stills in their full size within FCP. This would have been incredibly taxing on the system and generally overkill, even for an HD job. I created Photoshop actions to automate the conversion of the stills. The 152 5Dmk2 JPEG stills were converted from 5616×3744 to 3500×2333. The stills from the G10 come in a 4:3 aspect ratio (4416×3312) and were intended to be used as black-and-white portrait shots. Another Photoshop action made quick work of downsampling these to 3000×2250 and also converting them to black-and-white. Photoshop CS4 has a nice black-and-white adjustment tool, which generates slightly more pleasing results than a simple desaturation. These images were further cropped to 16:9 inside FCP during the edit.
Frame from My Room video
Once I had completed these conversions, the edit was pretty straightforward. The project was like any other PAL-based HD job (1920×1080, 25fps, ProRes). The Canon 5D creates files that are actually easier for an editor to deal with than RED, P2 or EX files. Naming follows the same convention as most what DSLRs use for stills, with files names such as MVI_0240.mov. There is no in-camera SMPTE timecode and all imported clips start from zero. File organization over a larger project would require a definite process, but on the other hand, you aren’t fighting something being done for you by the camera! There are no cryptic file names and copying the files from the card to other storage is as simple as any other QuickTime file. There is also no P2-style folder hierarchy to maintain, since the media is not MXF-based.
Singular Software and Glue Tools are both developing FCP-related add-ons to deal with native camera files from the Canon 5D. Singular offers an Easy Set-up for the camera files, whereas Glue Tools has announced a Log and Transfer plug-in. The latter will take the metadata from the file and apply the memory card ID number as a reel name. It uses the camera’s time-of-day stamp as a timecode starting point and interpolates clip timecode for the file. Thus, all clips in a 24-hour period would have a unique SMPTE timecode value, as long as they are imported using Log and Transfer.
Frame from My Room video
My final FCP sequence was graded in Apple Color. Not really because I had to, but rather to see how the footage would react. Canon positioned the 5Dmk2 in that niche between the high-end amateur and the entry level professional photographer, so it tends to have more automatic control than most pros would like. In fact, a recent firmware update added back some manual exposure control. In general, the camera tends to make good-looking images with rich saturation and contrast. Not necessarily ideal for grading, but Stu at ProLost offers this advice. Nevertheless, I really didn’t have any shots that presented major problems – especially given the nature of this shoot, which was closer to a documentary than a commercial shoot. I could have easily graded this with my standard “witches brew” of FCP plug-ins, but the roundtrip through Color was flawless.
As a first time out with the Canon EOS 5D Mark II, I think the results were pretty successful (click here to view). I certainly didn’t see any major compression artifacts to speak of and although the footage wasn’t immune from the “jello” effect, I don’t think it got in the way of the emotion we were trying to convey. A filmmaker who was serious about using this as the principal camera on a project could certainly deliver results on par with far more expensive HD cameras. To do that successfully, a) they would need to invest in some of the rigs and accessories needed to utilize the camera in a motion picture environment; and b) they would need to shoot carefully and adhere to set-ups that steer away from some of the problems.
What about 24fps?
25fps worked for us, but until Canon adds 24fps to the 5Dmk2 or a successor, filmmakers will continue to clamor for ways to get 24p footage out of the camera. Philip Bloom and others have posted innovative post “recipes” to achieve this.
I tested one of these solutions on my cut and was amazed at the results. If I needed to maintain sync dialogue on a project, yet wanted the “film look” of 24fps, this is the method I would use. It’s based on Bloom’s blog post (watch his tutorial video). Here are the steps if you are cutting with Final Cut Pro:
1. Edit your video at the native 30fps camera speed.
(Write down the accurate sequence duration in FCP.)
2. Export a self-contained QuickTime file.
3. Conform that exported file to 23.98fps in Cinema Tools.
(This will result in a longer, slowed down file.)
4. Bring the file into Compressor and create and apply a setting to convert the file, but leave the target frame rate at 23.98fps (or same as current file).
5. Click the applied setting to modify it in the Inspector window.
6. Enable Frame Controls and change the duration from “100% of source” to a new duration. Enter the exact original duration of the 30fps sequence (step 1). (Best results are achieved – but with the longest render times – when Rate Conversion is set to “Best – high quality motion compensated”.)
7. Import the converted file into FCP and edit it to a 23.98 fps timeline. This should match perfectly to a mixed version of the audio from the original 30fps sequence.
I was able to achieve a perfect conversion from 30fps to 23.98fps using these steps. There were no obvious optical flow artifacts or frame blending. This utilizes Compressor’s standards conversion technology, so even edited cuts in the self-contained file stayed clean without blending. Of course, your mileage may vary.
The edited video segment was 1:44 at 30fps and 2:10 at the slower 23.98fps rate. The retiming conversion necessary to get back to a 1:44-long 23.98 file took two hours on my MacBook Pro. This would be time-prohibitive if you wanted to process all of the raw footage first. Using it only on an edited piece definitely takes away the pain and leaves you with excellent results.
Cameras like the Canon EOS 5D Mark II are just the beginning of this DSMC journey. I don’t think Canon realized what they had until the buzz started. I’m sure you’ll soon see more of these cameras from Canon and Nikon, not to mention Panasonic and even Sony, too. Once RED finally starts shipping Scarlet, it will be interesting to see whether this concept really has legs. In any case, from an editor’s perspective, these formats aren’t your tape of old, but they also shouldn’t be feared.
©2009 Oliver Peters