As I’ve mentioned before, we shot the documentary Homeschool Dropouts on the 5DmkII in August, and posted it during September. It was a great learning experience, since it was our first time shooting video on a dSLR. Below is the worst shot from the project – all of the 5D’s image issues are visible in it. All of them can be avoided in-camera and all but one of them can be repaired in post (not counting the awkward composition).
Above is the final image as it was rendered from After Effects. Firstly, we have repaired the exposure. This was a very early shoot, before we started using the Magic Lantern firmware, and without its live histograms and zebra bars, getting the right exposure was tricky. Even though the camera only saves an 8-bit image, there is lots of room for correction, and since it comes from a 14-bit sensor, there is a surprising amount of latitude recorded.
Next, I adjusted the color. The green cast on the top of the wall is actually sunlight bouncing off the lawn in front of the porch. Since I had over-exposed a lot, and eyeballed the white balance badly, slight color changes like that were amplified, but it’s a great testament to the color sensitivity of the camera that it picked that up so vividly.
This shot also required a little bit of denoising. Even though we were using a low ISO setting, I had enabled Highlight Tone Priority on the camera. There’s some dispute as to how useful this setting is for video, and while it does provide more latitude in the highlights, it also adds some strange shadow noise. I used it on several early shoots, but I’m a little more leery of it now. I would use it with caution.
There’s also some subtle moiré pattern on Mr. Swanson’s sleeve. This is the Achilles heel of all of Canon’s video-enabled dSLRs, and it can be tricky to spot on the viewfinder. It’s also not especially predictable; note how it appears on the invisible pattern of the oxford cloth shirt, but not at all on the very pronounced weave of my jacket.
This cannot be fixed in post, but slightly adjusting the camera’s position, zoom, and/or focus will often make obvious aliasing and moiré artifacts vanish. I overlooked this issue, like all the others in the shot, because the camera was new, the shoot was hasty, and I was on the wrong side of the lens. Still, no excuse.
But enough dwelling on the worst shot; have a look at some of the better footage that we got:
All the outdoor shots were natural light with a single reflector, and the indoor shots were using available lights in the various homes rather than a professional light kit. Outdoor shots generally used Canon’s EF 28-135mm zoom lens, and most interviews used EF 50mm 1.8 primes.
Minor grading was applied to each shot, but it was extremely limited since the raw footage was so good. Some interviews got a subtle vignette, and there was a bit of levels work here and there, but most of these shots are pretty natural. I’ve downsized all the 1080p screenshots to 720 for bandwidth reasons, but it’s still big enough to see noise, any artifacts and also the sharpness that the camera is capable of.
I forgot to mention audio, or how we actually got footage from the camera into the edit. This is an important part of posting, so here’s our process:
The Canon 5DmkII records to MPEG4 files at about 48mb/s. For reference, HDV is MPEG2 at 25mb/s. MPEG4 is a far more efficient codec, so I figure that there’s actually more than twice the image data contained in twice the bitrate, but it’s not a good editing codec, so we converted everything to Cineform as we pulled it onto the computer.
The 5D also shoots 30 frames per second, which is a problem since NTSC video actually runs at 29.97 frames per second. Fortunately, Cineform automatically fixes this, conforming to the proper framerate and stretching the audio that extra 0.03 fps during conversion. We never had any sync issues once we realized we also had to stretch our external audio as well.
The interviews were recorded with a Sennheiser 100 G2 wireless mic running into a Zoom H2 audio recorder. All of the Botkin standups were recorded with that same mic running directly into the camera. Being primarily a stills camera, the 5D has crummy audio preamps, and they are by default set way too high and on automatic gain.
Using Magic Lantern, we were able to manually set the analog and digital gain at the appropriate levels, 0db and 6db respectively. With the Sennheiser receiver cranked all the way up to -6db, the audio signal was hot enough to need no real in-camera amplification, and so we got a very clean signal.