Canon EOS 1D Mark IV Perspective
The Canon EOS 1D Mark IV is a technological tour de force, in many ways unrivaled in its class. The Mark IV features a 16.1-megapixel CMOS imager, a new 45-point autofocus system, and a better-than-average low noise floor for capturing extraordinary still and moving images in low light. The new Mark IV is well-balanced and comfortable in the hand, and its all-magnesium body is rugged as hell. In short, this is one impressive still-image capture machine.
The evolution of the DSLR as a video-capture device coincides with the evolution of the still shooter, who increasingly provides, as a matter of course, client-requested video for the Web. This change in the still photographer's job description has precipitated significant changes in the DSLR itself. A highly efficient video capability has migrated upward to the major manufacturers' flagship cameras, such as Canon's EOS 5D Mark II, EOS 7D, and EOS 1D Mark IV, and Panasonic's Lumix DMC-GH1. For this class of user, certain challenging aspects of the cameras' usability, practicality, and performance—including aliasing artifacts and obvious macroblocking—are not deal-breakers. For clients of still photographers who want HD video delivered along with the photography, they generally will use video on the Web only, not for broadcast or big-screen projection.
Yet for many of us dwelling primarily on the motion side of the setup menus, the question remains: Can the Mark IV (and similar HDSLRs) make the grade as an ersatz high-quality HD camera for low-budget feature films, commercials, and compelling art pieces (such as Philip Bloom's visual poem Prague)?
There are many limitations inherent to a still camera operating in the far more demanding video environment. The Mark IV cannot and should not be regarded, for most applications anyway, as a replacement for your Sony PMW-EX1, Panasonic AG-HPX170, or Canon XH G1. It is simply another arrow in your video shooter's quiver, to be exploited like any other tool. In this case, it's within the parameters of what a still camera, albeit a superb one, can be reasonably expected to offer in the video realm.
The Mark IV's primary competitor is the Nikon D3s, which shoots 720p only and uses motion JPEG compression. Like most HDSLRs, the Mark IV employs a modern, web-focused H.264 encoder, the merits of which for broad video application I'll discuss in detail. While the Mark IV offers more resolution than the D3s in Movie mode, my tests suggest the Nikon offers considerably better noise resolution than the Canon, especially at 2500 ISO and greater.
The Mark IV offers HD capture at 1920x1080 (1080p) at 29.97fps, 25fps, or 23.98fps or 1280x720 pixels (720p) at 59.94fps or 50fps. While the built-in camera mic records in mono only, an external 1/8in. mini-jack allows stereo recording and the use of a professional microphone if desired. I'm hardly a fan of the flimsy 1/8in. input plug, which can easily fail or become intermittent in the course of a shoot. Suffice it to say, this maddeningly unreliable plug is one of the many compromises of using a low-cost HDSLR for serious video production. Look for sturdier, properly balanced XLR connectors in future HDSLR models. External audio capture remains an alternative to relying on the Mark IV's audio capabilities, of course.
The ergonomics of the Mark IV are superb for still-photo applications. For video use, different requirements apply. The usability and performance implications of the size and shape of the HDSLR have a profound effect on the life and livelihood of the video shooter.
For one thing, the operating buttons on the camera body are too small and haphazardly arranged for my liking. The camera lacks proper tactile buttons, the viewing screen in Live View is difficult to see at an angle, there is no built-in timecode provision, and finding and keeping critical focus is an exercise in frustration.
HDSLR lenses, including the popular Canon 24-105mm EF, were not intended to hold focus while zooming or to execute precise, repeatable camera moves where the ability to follow focus is critical. These lenses might be described more accurately as variable focal length lenses. They are not true zooms as video shooters have come to understand the term within the film and video world. Follow focus can be difficult or impossible to pull off, even with proper monitoring. Indeed, the mechanics of leaning against or even touching the body of these lenses while recording can lead to distracting vibration and often wild encoding artifacts on screen.
For many filmmakers, the key attraction of the HDSLR is the inherently narrow depth of field because of the large sensor and long lenses required to produce a "normal" field of view. The large sensor also serves to reduce contrast-robbing diffraction, a shortcoming of all lenses at small apertures that results in a significant loss of sharpness. In videocameras with tiny imagers of 1/3in. or less, diffraction artifacts often begin earlier in the f/stop range. Shooting with 1/3in. or 1/4in. cameras stopped down beyond f/4 or f/5.6 is never recommended, as it can result in images that are effectively no better than standard definition in many cases.
Curiously, the Mark IV offers a more convoluted video control implementation than the 7D. The 7D, which has a dedicated switch to change shooting modes, is much more convenient. Once Movie mode is engaged, a simple press of the Start button initiates video recording.
With the Mark IV, once Movie mode is engaged the shooter must select the proper Live View mode then locate the microscopic FEL button to initiate recording. With gloved hands or in the frenzy of a harried shoot, reaching for a button slightly bigger than a pinhead can be extremely inconvenient. In Movie mode, the Mark IV allows manual setting of key camera functions, including shutter, aperture, and ISO. This is certainly a valuable improvement over earlier HDSLRs, such as the original incarnation of the 5D Mark II that controlled such key parameters automatically. (Manual controls in the 5D Mark II were added recently via a firmware update.)
One key advantage of CMOS sensors used in HDSLRs like the Mark IV is the ability to address each pixel individually for maximum creative and technical control. CCD sensors, in contrast, offer superior performance overall but must gang their pixels as a group and clock them in a more organized way. Indeed, the HDSLR is able to derive a 1920x1080 image from a native 4K sensor simply by cropping out the middle of the frame. One has to wonder what the reaction of shooters might be if Panasonic or Sony used such a tactic in their professional videocameras.
The truth is any camera that shoots progressive still images is capable of capturing full-frame video in some shape or form. The challenge for HDSLRs, however, is maintaining satisfactory video quality, owing to the lack of computational horsepower and diminished ability to offload data quickly from the CMOS sensor to the processor and encoder, and ultimately to the flash memory recording medium.
Prior to the introduction of the EX series, Sony was said to have toyed with the idea of recording the MXF volume to inexpensive SD memory, only to return to the much faster, more reliable SxS media, despite the higher costs involved. The reliable capture of HD video requires the recording media to handle the data load without dropping frames or producing incomplete frames.
The inconsistency in speed and quality of consumer-grade CompactFlash or SD flash memory can cause problems for the capture of high-bandwidth video. Yes, there is a reason Panasonic P2 cards are built to military specifications and use four hand-selected flash-memory chips in a RAID 0 configuration: ensuring high-speed reliability when writing 100Mbps files to solid-state media.
The offload speed of data from the HDSLR sensor is not critical in still photography. The camera processor effectively has all the time in the world to complete its task, even with the application of substantial noise reduction. In Movie mode, however, it's a different kettle of bits. The (relatively) lethargic offload of data poses many challenges that tend to work against the capture of crisp, artifact-free HD images.
In dedicated video cameras such as Sony's HXR-NX5U, data is pulled off the sensor array pixel by pixel from multiple directions and subsequently processed and encoded. In HDSLRs such as the Mark IV, the readout of data is performed one row of pixels at a time at the side of the sensor array. The reading of data in this way from left to right is not only slow, but the HD frame's vertical resolution is also compromised because only every other row is sampled to reduce the processing load. This economy substantially increases the risk of aliasing artifacts, as the missing vertical line must be interpolated—with all the inaccuracies that that entails. This loss of resolution in Movie mode is in addition to whatever cropping is required to achieve the desired 16:9 Movie frame.
In general, the number of output channels from the sensor array largely determines how much and how fast the picture samples can be processed. This directly affects the resolution and integrity of the Mark IV's 1920x1080 or 1280x720 movie files at 24fps, 30fps, or 60fps.
Compared to true video cameras, the relatively slow offload of data from the HDSLR sensor has another undesirable effect: the dramatic increase in skew and rolling shutter artifacts, which are exacerbated by the long scan times required to complete each frame.
Needless to say, the data load for a typical HD frame with moderate compression is daunting. In Sony EX camcorders, more than 2 million lines of code are required to process the data from every pixel on every line from a 1/2in. imager. The Mark IV, with a sensor many times larger than those of the Sony EX series, faces a monumental challenge if anything close to a comparable scan or sample rate were employed.
The Mark IV's encoder is necessarily limited by a processor that's optimized for still images. Like most HDSLR models, the Mark IV records heavily compressed H.264 video. The popular MPEG-4 variant promulgated by Apple is efficient for web streaming at low bit rates, but it is far less effective at high-bit and high-compression rates like those encountered in the HDSLR. Unlike the familiar video codecs—AVC-Intra, AVCHD, XDCAM EX, etc.—there is no established standard for H.264 video encoding; indeed, HD quality can be all over the map, and under challenging, high-motion conditions, it often is.
So, if computational prowess is the limiting factor for high-quality HD acquisition, why not simply beef up the processing and use a more appropriate video codec in the HDSLR? Why not sample every pixel in every line and/or dramatically increase the number of outputs from the CMOS imager? This is what is done in videocameras, so why not do it here? It would certainly reduce artifacts, produce markedly better HD video, and go a long way toward bridging the performance gap between the HDSLR and a true video capture device.
The problem is, there are compromises to consider. Dedicated videocameras with much smaller sensors and pixel-by-pixel processing, such as Sony's new HXR-NX5U, feature large heat sinks that divert the processor's heat away from the imager and the lens mount cavity. The HDSLR's small size (as demanded by professional still photographers) and lack of large heat sinks impose severe constraints on the amount of data that can be reasonably collected and processed, which leads inevitably to significant compromises in the integrity of the HD image. This is in addition to the limitations of the flash memory itself, which imposes its own set of constraints regarding data throughput and consistency of write speed.
With so many choices in HD cameras, it has become something of the shooter's mantra to seek the best tool for the job. Yes, the HDSLR in general, and the Canon EOS 1D Mark IV specifically, has a place in the hands of serious shooters. There will always be projects for the Web or small screen that can benefit handsomely from the camera's unobtrusiveness, its low-light sensitivity, and its dynamic response. Perhaps you're shooting a music video, a stylized commercial, or a tightly controlled visual poem such as Philip Bloom's Prague. The HDSLR camera can work very well in these applications.
For most high-value productions, documentaries, routine video applications, and most run-and-gun reality TV, however, the Mark IV is not going to replace your Sony EX1 or F23 or Panasonic AG-HPX300 just yet. Even in the face of Canon's enormously successful HDSLR models, the company has seen fit to introduce a new 50Mbps 4:2:2 codec for its latest-generation videocameras.