Did Canon release a new camera recently, or something? A large part of the world is talking about an upgrade of the 5D, whereas the company itself is more interested in selling us a competitor to the Sony F3, and even admitting to some ambition that it’ll take some of what might otherwise have been Alexa or Epic jobs. Regardless of how realistic that seems, it does highlight Canon’s hopes for the camera, and goes some way to explaining why the original price announcement seemed a little at odds with what people were expecting by way of a DSLR without the problems.
I think that a reasoned opinion of the camera depends largely on whether you insist on trying to interpret it as an upgraded DSLR, or whether you’re willing to accept Canon’s line that it’s more of a competitor to the Sony F3. Visual Impact‘s recent open house in London certainly pushed the latter, demonstrating the C300 directly against an F3 and, daringly, an Alexa. The paper spec of the C300 has been the subject of much discussion, but this was a chance to look at the pictures on some really nice displays and make a more subjective assessment.
The C300 already has an enviable reputation for producing pictures that are subjectively very nice, and I am pleased to be able to confirm it. Canon’s pedigree as a sensor foundry is already well-established in the stills market so it is perhaps not surprising that, having taken the decision to bend their R&D efforts toward a motion picture sensor, they’ve produced something rather nice. I must qualify this; tales of being able to casually select 20,000 ISO sensitivity without fear of noise are apocryphal, and even though the noise characteristic is not unpleasant, the codec can struggle with the sheer amount of entropy in the image under these conditions. That said, speeds up to a couple of thousand do not cause serious problems.
It’s worth explaining why this matters, and while I’m sure the C300 will be a great documentary camera, the purpose of all this speed isn’t necessarily so we can shoot available-light night exteriors of a black cat in a sack. Sensitivity really means low noise, inasmuch as what would have been the lower end of the luminance range becomes usable picture information. If that range is reasonably low in noise, we can place more of our picture information there, leaving more information for encoding highlights – and decent highlight response is a big deal in video cameras. It is for this reason that people shoot electronic cameras near to the upper end of their sensitivity limits; it’s a simple matter of trading noise for dynamic range.
The advantage that Canon has over Red, whose competitively-priced Scarlet looks much better on paper, is that Canon has a nearly-4K sensor that they’re decoding into a really nice HD image, whereas Red has a nearly-5K sensor that they’re usually decoding into, from what I’ve seen, a pretty average 4K image with all the compromises that entails. This is of course merely a design choice, but I applaud Canon’s technical honesty in not trying to make a 4K RGB camera out of a 4K bayer sensor. Arri do something similar with Alexa, with similarly smooth and low-noise results. I suspect this is responsible for the C300’s reputation, which I’m pleased to confirm, for “tight” noise that is somewhat organic in character. It’s certainly much less objectionable than the blobs of snowiness that are often visible on cameras that try to decode a bayer sensor to an RGB image of the same pixel dimensions.
Canon has not allowed, or at least not allowed to be published, dynamic range tests, but it is reasonable to assume that they will do well here, especially at higher ISOs.
This is far from the whole story, though. My principal technical concerns about the C300’s spec sheet were that it didn’t have 10-bit output, and that the 50Mbps MPEG-2 codec was likely to be inadequate for at least some scenes. I hasten to add that I don’t think a camera like this needs 10-bit output for any good technical reason. Few sensors, even sensors as quiet and as nicely oversampled as the C300’s, actually have more than 8 bits of real, noise-free information, especially with logarithmic encoding. I think it may still need 10-bit because that is what the industry is used to assuming is required, and, regrettably, the people who make those sorts of decisions often don’t understand sampling theory very well. So, that’s either a problem if you’re a producer who doesn’t know how to count, or not a problem if you’re a producer-director-cameraman who’s in full control of the production and able to make decisions based on reality rather than politics.
What is a more of a problem is the codec. I’ve written before about MPEG-2 as an acquisition tool and the tradeoffs inherent in any scheme that tries to use the similarities between frames to achieve compression. Briefly put, it’s mathematically hard work, and can be done either well or badly. Doing it well in real time is expensive, power hungry, and large, all of which are constrained factors on something like the C300. Throwing more bitrate at the problem helps only to an extent, because the extra compute horsepower required to encode the extra bitrate could just as well have been applied to doing a better job with the bits we had already.
The C300 also uses a different codec to the D-series DSLRs, if that’s your point of reference. This is a slightly more understandable engineering choice: more advanced codecs such as h.264 exacerbate the math problems, so while the cleverer technique may seem better on paper, the ability of a small, relatively low-cost handheld device that runs for hours on one battery to do it well is in question. Put simply, h.264 is not necessarily better than MPEG-2, depending on the circumstances, so I don’t necessarily criticize the choice of codec. Presumably, the C300 uses the same codec hardware as Canon’s XF series handycams, which are widely used for broadcast work.
The issue with all this is that 50 megabits of MPEG-2, encoded with the sort of precision that’s achievable in a handheld camcorder, is not really enough. It looks very good indeed in low-noise scenes, but if you choose to use that sensitive, low-noise sensor to its limits, there is visible codec smear and some motion artifacting in areas of noise. It’s a sort of swimming around of the noise artefacts, as if viewed through water. It’s an organic artifact, not quilting (“blockiness”), and it could be worse, but I think an external uncompressed or high-bitrate ProRes or DNxHD recorder would be worthwhile.
All these imaging impressions come through viewing the images encoded as ProRes on one of Sony’s new Trimaster OLED displays. These really are spectacularly good – at last, a display technology with no caveats beyond a terrifically high price (around UK£20,000 for the 25-inch type) and alarmingly short life of the OLED panel. With any luck, we’ll all be able to have these on our desks within the next few years, once they get the lifespan issues sorted out and start mass producing them. The image on these monitors was so good, so bright and punchy, so clean in the blacks, so free of artifice and sensitivity to viewing angle, that enthusiastic discussion of their excellence threatened to overshadow the C300 which had actually supplied the images.
The C300 has some nice ancillary features. There are two card slots, which take ubiquitous compactflash cards which are easily available worldwide. These two slots can be recorded simultaneously, creating an in-camera dual backup that makes the entire data-handling procedure easier and more reliable. There are all the usual inputs, including timecode, although the power input is 8.4V and on a flimsy domestic-style DC connector. The camera will render a waveform display and this display is of high quality, not some low-res histogram-style thing as we have to endure on DSLRs. It has the usual domestic style design problems, being made of curved plastic surfaces that are liberally smothered in buttons on all sides, but it’s no worse than many other things and I’m sure some people are already in the machine shop prototyping accessories in a variety of unnecessarily vibrant anodized finishes.
I still don’t think it’s worth anything like $20,000 U.S., in a world where you can buy a Sony F3 with better numbers. Happily, Canon seem to have realized it isn’t going to sell terribly well at that price either, so we’ll have to wait and see. I’m interested to see the dynamic range charts.