Grading Low Light DSLR Footage

Yeah, I know this has been done a million times before but I just got my Canon EOS 7D and wanted to test the low-light performance myself. I aslo wanted to see how well the footage could be graded. and while there are myriad examples of this sort of thing on the internet, when I went looking I didn't find one that showed me this in a way I wanted to see it. So who knows, maybe this will one day help someone out. 

A very important factor in evaluating the footage of any camera or codec is the ease of use. As much as I love getting down and dirty with color grading, I don't want to spend forever. The gold standard is film. You shoot film, it looks great. For decades the way to control the color of film was by adjusting printer lights to affect the transmission of 3 colors. Simple. The more you need to massage a digital image to get an acceptable image out of it, the more that digital system can be considered a failure. Yes, I know - cheap, fast, good; choose any two - if you're on a budget you can expect to spend more time achieving that good image. But you should still evaluate a system by how painful that sacrifice is.

So I kept this test simple. At my friend (artist, writer, and director) Sketkh William's birthday party, I grabbed some footage on the 7D. Barcade in Brooklyn is lit primarily by these novel little bag lights on the table. They have what can't be more than 20W yellowish tungsten bulbs - kind of like Christmas lights - stuffed inside brown paper lunch bags. Honestly it looks like there's a candle inside and you have to look for yourself because it's easy to feel as though there's a fire hazard. There is also dim light from the arcade game screens (it's really a cool place if you like classic games and/or craft brews) and very dim very sparse lights on the wall. Overall it's a pretty dark place. I took some shots at 1250, 2500, and 5000 ISO. The camera really can see in the dark, but you could see the noise in every image as well. I like noise when it's pleasant though, as long as it doesn't deviate too far from the look of the rest of whatever you're shooting.

For post, I converted the images to ProRes4444 using 5DtoRGB. The footage from the camera has 4:2:0 chroma subsampling and although converting it obviously won't add any color depth, any color grading application will have an easier time handling, and be able to perform more nuanced operations on, an extracted RGB image. To keep things really simple, I graded the images by eye in After Effects using only Colorista. Normally you'd want to use scopes and get precise, but I'll be doing a more in depth grading test next week with the new Technicolor color profile. For the time being, I just wanted a simple test with low-light footage. So first, here are some screen grabs from the original footage:

And here is the graded footage:

Overall I'm pretty happy. I wish I could show the actual quicktime file I output but I had to transcode it for Vimeo. In the actual file you can see the grain structure very well, but the above video will give you a pretty good idea. Here are some basic observations:

• 1250 was a nice setting but underexposed. I really like how the colors graded, but I had to boost the gamma and that created noise. 
• 2500 had a surprisingly acceptable noise profile off the camera and I didn't have to grade it very agressively. The noise is mostly noise at the sensor level which is more pleasant and while the 1250 footage started off cleaner, this is the better graded image because the noise is more natural.
• 5000 was overexposed. You can see the bag lights blow out. Yes, the faces are better lit here, but the overall image is brighter than it should be. In the graded image, I tried to supress the gain as much as reasonable to handle some of this.The noise, like with the footage shot at 2500 is camera noise and overall not horribly objectionable but still much more visible than I'd like. This setting fell into the typical pitfalls of overexposing a digital image.

2500 ISO was the best setting here and really the right setting for the environment. At 5000 the faces of are better lit and are less red in the native image, so while 5000 is great in those terms it creates more grain than I want and requires more grading. If I had any control over the environment I'd agument the bag lights and bring the exposure on people's faces up  to shoot at a lower ISO, but the point here was to grade an image from a poorly lit evironment. I think 2500 produced the best overall image (I took 3 shots at each setting and the ones not being displayed here suggest this as well).

This shows the importance of getting things right in the camera. These cameras don't have a whole lot of dynamic range to begin with so, much like reversal film, a little over or under can kill you. Under normal circumstances I wouldn't want to shoot at any of these ISO settings, but sometimes you have to. When this is the case, you have to accept that you're going to get some noise; but you have to understand that the best way to combat noise and a limited latitude is to expose properly. This will make grading easier and minimize the visual sacrifice that you're making. If you're lucky, instead of ruining your picture you'll add to its character.

Camera Test

The Camera

Since I announced the 52 Weeks Project my mind has been occupied over which camera to use. My first thought, and my wishful thinking, had me immediately considering a refurbished camera from Pro8mm or Du-All camera that I knew would be well-serviced and modified with crystal sync. A camera running crystal sync would allow location sound and that seems like a good thing at first. But, Super 8 cameras tend to be a bit noisome and, given the constraint of straight-8 shooting there's enough that can go wrong in a take, especially with sound, that considering I only get one take, just doing all post in audio seems smarter. I'll record reference sound of course, but dialog will be ADR. Those cameras are also expensive. They're probably worth it, but I don't really have the excess to spend.

So I started looking at reasonably-priced but solid cameras. I decided to save money by looking primarily on ebay and doing my homework on the cameras. there were a couple of must-haves in a candidate:

• Manual Exposure - I plan to rely heavily on fast emulsions, like Kodak's Vision 3 500T, which weren't available when these cameras were made and so cameras with only auto exposure won't work right.
• A good zoom range with decent optics - I generally prefer shooting with prime lenses, but that's not so easy to do in the realm of consumer cameras, which Super 8 cameras are, albeit of another generation.
• 24 Frames Per Second - A lot of these cameras were manufactured in Europe and have 25fps but not 24. Something to be wary of. A note here - you can get away with shooting at 18fps and make a cartridge stretch. Generally I won't be doing that; mainly because it increases the cost of telecine and I'm trying to work out a package deal based on the mainstay which is 24.
• Single-Frame Exposure - Most of these aren't built for automated time lapse but a lot of them were built to do single frame for animation and whatnot. Over the course of 52 weeks I may get bored and decide to do a stop-motion holiday special, or decide I'm the next Ray Harryhausen. More likely I'll decide on a manual time-lapse for a film. In either case I'd like the option.
• A useful filter threading - Since I'll be using fast film to cover my indoor bases I'll need to rely on neutral density filters in daylight situations. While I may at some point endeavor to rig a mattebox or something, I'll probably start off with circular screw-on filters.

The camera I was first drawn to, the Leicina Special, is a beautiful camera but doesn't have manual exposure. I began looking at the Canon 1014, which is nice but doesn't really justify the price increase over its little brother. So I started looking at the Canon 814XL-S which is a very well spoken-of camera and, being from the early 80s, a relatively new model. In the end, I compromised on a deal I couldn't pass up - a slightly older Canon 814 Autozoom that I found for under $100. I've shot a little with the 814 before and, assuming the one I just bought is in good condition, it's a really good camera; not the best Super 8 out there, but it meets my needs and should hold up to the task before it.

I should receive it next week and I'll run a test roll through it. As soon as I get that back I'll be posting it here, unless it turns out to be total trash. Until then, I'll be continuing to look at labs to handle the project; and that should be the topic of my next post.

(photo courtesy of flickr user Callum Rice)

Another Video SLR Convert

I personally have very mixed feelings about the rise of video DSLR cameras such as the Canon 5D mark II. I can't deny that some of the work being done on these things is amazing, and the Zacuto Shootout that's being released bit-by-bit is clear evidence of the potential of these cameras. This post, though, isn't about my opinions about this trend - that's coming soon. This is about promoting a friend and his work. Richard Crook, a DP from Dallas that I've worked with recently shot a short film on his 5D rig. I saw production stills this morning and while my first thought was "et tu, Richard?" my second thought was "well, it's certainly a step up from his HVX200/Redrock rig (sorry, Rich, you know how I feel about that kit). This afternoon Rich caught me online and sent me a link to a preview of "The Key," and I have to say, it looks fantastic. With some lightweight dolly and crane gear, a good video rig, and a small crew, he shot something that is beautiful, professional looking, and at a value well beyond its budget.



After watching that, I wanted to share Rich's blog on the tools he used to turn his 5D into a real cinematic tool. So, please take the time to check out his "Complete Idiot's Guide to the Canon 5D MkII."