CineStyle for Canon EOS DSLRs
- Why use Cinestyle?
- Video noise in dark regions
- What about taking stills?
- Are there any disadvantages?
- Installing Cinestyle
- Setting up the Canon 5DMkII
- Shooting with Cinestyle
- Post processing the footage
- Working with LUTs
- What about Linear Colour Space?
- Where do we go from here?
- A final plea
If you haven’t seen it already, Technicolor have released a custom Picture Style called Cinestyle for the Canon DSLRs.
Apparently Technicolor have been working on this for the last year in an attempt to bring the Canon DSLRs more in line with other digital film cameras such as RED and the Alexa. The Cinestyle profile produces a very flat looking footage that retains a lot more detail in the darker areas. Technically speaking, it encodes the images received from the sensor into log colour space. Anyone used to working with raw footage from film will recognise the flat grade look to the footage. The aim is that it produces footage that is much easier to colour correct while retaining as much quality in the image as possible.
This post is an attempt to try to collate and filter as much up to date and correct information about Cinestyle and the practicalities of using it with the 5D MkII. There seems to be a lot of misunderstanding out there about what this is all about, particularly when it comes to working with lookup tables, or LUTs. (Nobody’s perfect though, so please leave a comment if you think I’ve got something wrong myself!)
Much of this information I’ll be sourcing from this thread on the Cinema5D forum. I’ll try to provide links as I go through to give proper credit and so that you can read the posts for yourself. If there are any dead links, please let me know in the comments.
Why use Cinestyle?
So first thing’s first; what are the benefits of using Cinestyle? Well, shooting in log colour space means that you retain more information in the darker and brighter areas. Note, that this doesn’t mean that you’re suddenly achieving greater dynamic range. The chip on your camera hasn’t changed and so you can’t magic dynamic range out of nowhere. All it means in practice is that you are getting better detail in the dark and light areas within the existing dynamic range.
All this messing around with the colour space is done before the lossy stages in the camera i.e. H264 compression (Source: post by Matt Thomas) which, of course, makes sense as otherwise it would be further degrading the signal.
The thing worth noting about H264 compression is that it is most aggressive at the low and high ends of the luminosity range, so by using the log colour space, it is effectively moving the valuable colour information away from the nasty compression artifacts. Again, Matt Thomas explains the concept very well here. He says:
The principle behind CineStyle is to move your shadow details up, away from the more destructive shadow region, which the H264 CODEC will damage more aggressively. By moving your “blacks” up into the lower mids, you will have more compressed data devoted to subtle gradations from the h264 CODEC. The same is true for highlights. By keeping the contrast very low, it allows for your highlights to be moved further from the destructive part of the h264’s white points, which, like the black points, have much less detail allowed by the CODEC than mid tones allow.
So I decided to put it to the test.
Video noise In dark regions
The following images illustrate the big advantage of using Cinestyle with video. It’s designed to test the amount of noise in the dark levels of the image.
|The image on the right shows a still from the video output from the camera when Cinestyle is enabled. You can see, compared to the next set of images, the low contrast here produced by Technicolor’s picture style.||
Cinestyle picture style (no LUT)
Once the lookup table is applied (in the right hand image below), the result is very similar to the video produced by the Normal picture mode (below on the left). More info later in this post on how to apply the lookup table.
Normal picture style
Cinestyle picture style with LUT
This is where things get interesting. If we now apply a Exposure filter in After Effects with a value of +7.7 to both images, we get the following:
Normal with +7.7 Exposure
Cinestyle with LUT and +7.7 Exposure
You can see, even in the thumbnails, that there is noticeable banding in the image captured from the Normal picture style image (to see it more clearly, right click on each image and select “Open Link In New Window”). It’s pretty obvious that the Cinestyle video is giving us much more information in the dark areas.
What about taking stills?
Okay, this bit was a surprise… If you take a picture using the Cinestyle mode while using Jpeg as your storage medium, you get the flat low contrast appearance in the image. All fine so far. What is surprising however, is that if you set the 5DMkII to store RAW, there is still a subtle yet noticeable difference. Usually, changing picture mode (even switching to black and white Monochrome mode) has no effect on the RAW image at all. However, in this case, something interesting is happening with Cinestyle.
With the two images below, at first glance the RAW doesn’t appear to be particularly affected by Cinestyle. RAW files basically ignore most of the camera settings like picture profiles, white balance, etc., so this would be as expected.
Normal picture style
Cinestyle picture style
However, if we now overexpose both images, we get this:
It’s a very subtle difference which you probably won’t notice unless you view the full size images side by side (again, just right click on each image and select “Open Link In New Window”). If you look in the darker areas, you’ll notice that the Normal picture profile hue tends towards red in the darker areas, whereas Cinestyle appears to retain the correct hue through to the darkest point.
Wow. So Cinestyle seems to be doing something genuinely funky with the colour processing pipeline in the 5DMkII. Something that none of the other picture styles do. This only goes to confirm that Technicolor must have been working very closely with Canon to implement Cinestyle.
Are there any disadvantages?
Yes, possibly. In tests I’ve done, it’s quickly apparent that there’s a black offset. A large number of posts online suggest that black is mapped to a value of 16. In my test however, it only appeared to be around 8. This would seem to be confirmed by the Cinestyle LUT. Going into After Effects and making a solid layer at RGB <8,8,8> and applying the Cinestyle LUT (see later), the resulting RGB value that the LUT returns is <1,1,1>, whereas an input of RGB <7,7,7> results in <0,0,0>.
If you want to see the black offset for yourself, the test is very easy to do on your camera and is noticeable just looking at the LCD viewfinder. Just leave the lens cap on and take two short bursts of video, one with Cinestyle, and one in Normal mode. Press the play button and flick between them using the front dial, you’ll be able to see that the black is obviously lighter in the Cinestyle footage. When I tested the colour values in After Effects, it gave values around 7 or 8 as the minimum.
Similar tests with overexposure showed that the white was not offset, i.e. with full exposure, it used the maximum white value of 255. The offset in the black helps to avoid the more destructive compression of H264, but it also means that we’re losing colour resolution. This could result in banding issues if the colour grading is pushed too far in the mid tones where there might not have been issue before. For a good example of this, see this post by Richard Crook. Also, check out this video which discusses some of the problems involved using flatter colour profiles.
Which then leaves us in a bit of a dilemma. Should we be using Cinestyle at all? Unfortunately, I don’t think there’s an easy answer. I’d say that it depends on what you’re going to shoot. As Richard says in a later post:
If you are shooting a situation like, for example, in a forest…then banding is a moot point because there is so much texture it doesn’t matter. But shooting a low dynamic range subject like the sky or a wall, both with subtle gradients, a flatter profile like Cinestyle isn’t a good choice.
That makes sense to me. From a post production point of view though, it’s a bit of a pain, because if the camera operator is switching between different colour profiles on a per shot basis, then that’s one more thing you (or the clapper loader) has to keep track of.
To install the Cinestyle profile, you first need to download it from Technicolor’s website.
Once you’ve got that, you need to make sure that you have the EOS Utility installed on your machine. It’s supplied on a CD that came with the 5DMkII, but if you don’t have it handy then you can download EOS Utility from Canon US. The only issue with downloading it, is that it’s engineered by Canon to be an update-only install (why do they make things so hard for us!!). Luckily, it’s quite easy to turn the update installation into a full one. Full instructions can be found on Keith Cooper’s photography website.
The step by step instructions for installation can be found at Technicolor, but for your convenience, I’ve copied them in below.
- Ensure that EOS Utility v2.6 or later is installed on your computer
- Download the Technicolor CineStyle Picture Style file
- Connect your camera to your computer using the appropriate USB cable
Note: for the EOS 5D Mark II you may need to set “Communication” to “PC connect.” in the camera’s menu
- Start the Canon EOS Utility
- Select “Camera settings/Remote shooting” on the main window
- Once the capture window opens, click the camera icon
- Click “Register User Defined style” under “Shooting menu”
- Once the new page “Register Picture Style File” opens, select one of User Def. 1, 2 or 3 at the top of the new page
- Click the open file icon
- Select the CineStyle.pf2 file you’ve just downloaded
- Click on OK
Setting up the Canon 5DMkII
There are a couple of subtle issues that you should be aware of to maximise your the quality of your footage. Neither of them are killers, but it’s a good idea to be aware of them.
The first thing to know (and the source for the following explanation is a great in-depth article at Josh Silfen’s website), is about how the 5DMkII works with the ISO setting. To paraphrase Josh’s article, the sensor in your camera has an intrinsic ISO setting associated with it of ISO 100. The ISOs that are at doubling sensitivities above this (i.e. 200, 400, 800, 1600, 3200, 6400) are also native ISO’s (derived, I suspect, from directly manipulating the way the sensor is employed). All the other ISOs are based on these and are either 1/3 stop over or under exposed.
For example, it is approximately true that:
ISO 320 = ISO 400 x 2^(-1/3)
ISO 500 = ISO 400 x 2^(+1/3)
This is a digital process after the image has been captured, and means that from a visual point of view, the pull down ISOs of 160, 320, 640, etc. will appear to have less noise. Unfortunately, the stopping down also means we have less dynamic range, as the highlights lose 1/3 of a stop as well. We end up with a trade off, and so it’s definitely something to be aware of. Please read Josh’s full article on this for more in depth information.
The other thing to know, is why you probably shouldn’t use Highlight Tone Priority. Again, Josh comes to our rescue with another brilliant post on this which you can find here. Essentially, the message is that Highlight Tone Priority doesn’t give you any extra dynamic range, and you have a little less control over exactly what the camera is capturing. You’re doing colour correction in-camera, rather than in post.
Shooting with Cinestyle
The main thing to know is not to set the exposure for your video with the CineStyle profile enabled. As Richard Crook notes in this post, going a stop either side of the correct exposure makes little difference to what you can see in the camera’s LCD screen. That’s a big change to be unaware of.
So unfortunately that means you’ll need to set up the exposure using the Normal picture style, and then switch to Cinestyle for capture. Not ideal. It’d be great if it were possible to capture using one style and preview using another, but I guess we’ll have to keep wishing.
Post processing the footage
Once you’ve taken your video, what’s the best way to process it? RareVision’s tool 5DToRGB will convert your H.264 quicktime movie into DPX files using some sort of clever processing technique to achieve better results than the standard quicktime codec for decompression.
If you’ve got a lot of files to convert, you can use this batch tool from Noside to convert multiple videos at once using 5DToRGB.
The footage that comes out of the camera looks very flat. That’s why Technicolor provide a lookup table (LUT) to view the footage with. In order to use it with After Effects, you can download for free Magic Bullet LUT Buddy 1.0. This allows you to apply the LUT to a layer as an effect. There seems to be a couple of issues with LUT Buddy crashing, but it is possible to work around the problem by converting the LUT to a different format. More information on this fix from Lars Steenhoff’s blog post.
Working with LUTs
There seems to be a fair bit of misunderstanding in the online DSLR community about what LUTs are and how to work with them. It’s understandable, as they’re not something most photographers have to deal with, so there follows a brief explanation.
In an ideal world, before manipulation in post with CGI and compositing, all footage should be converted into linear colour space (sometimes also called “scene referred” colour space). Linear colour space is where the brightness of a pixel is directly proportional to the amount of light that it received in the camera, i.e. a doubling of pixel intensity, means double the received light.
Most people are used to using images that are encoded in sRGB colour space. This has a gamma correction curve of approximately 2.2 encoded into the intensities so that it displays correctly on a monitor (which has a response curve of 2.2). But this means that for an sRGB image, if a pixel receives double the intensity in camera, it won’t be double the brightness in the image.
The problem with using and manipulating sRGB images is that a lot of mathematical operations in compositing don’t work due to the maths being affected by the baked in gamma curve. On the other hand, although we can use linear images, they don’t look very good when directly viewed on monitors and it’s very hard to make aesthetic judgements.
This is where LUTs are a big help. They allow us to view images in one colour space (i.e. sRGB), while the software processes them in another (i.e. linear). It’s a little bit like looking through a pair of sunglasses; the image isn’t being being changed, but it’s altered at the last minute for your benefit so that you can see it properly.
So in VFX, LUTs are used all the time to work with footage in linear colour space during compositing. It means that the software can do all it’s operations correctly in linear space, while we can view the result through the LUT that turns it into something that we can view properly on our monitor. In addition, we’ll often encode a grade (colour correction) into the LUT so that we can see how the final image will appear once the colourists have done their magic on it and applied the final look.
The problem with software like After Effects, is that their colour space handling is extremely confusing to deal with. Seeing as After Effects doesn’t directly support LUTs (as pointed out in the comments section, CS5 now has a plugin to apply LUTs in a similar way to Magic Bullet, unfortunately there’s still no way to load a LUT into the colour management settings), then I’d generally recommend applying LUT Buddy to an adjustment layer at the top of the main composition. As you work, everything can be viewed with this applied.
Colour corrections can be done before or after the LUT layer. In fact, you can even throw away the LUT altogether and do your own grade from scratch. It’s a little bit arbitrary for reasons explained in the next section.
What about Linear Colour Space?
That’s a very good question. The big problem with Cinestyle is that the supplied LUT converts from a log based colour space to something very similar to the original REC 709 colour space, and NOT linear. A lot of people online are saying that the LUT does convert to linear, but it doesn’t as the final result still looks perceptually correct on a 2.2 monitor. This means that it must have a 2.2 gamma correction applied to it, and so it’s probably REC 709 or sRGB.
As of this moment in time, Technicolor have not released a LUT to go from Cinestyle to linear. This is a bit of a major issue for anyone wanting to composite with this footage. In theory, it should be possible to use the LUT to convert from Cinestyle to REC 709, and then convert from REC 709 to linear, although I’ve not tested it’s accuracy yet.
Where do we go from here?
In lieu of Technicolor releasing a Cinestyle to Linear lookup table, I believe the next step is to look into measuring the camera response curve and deriving my own lookup table to convert to linear space. I believe a paper exists by Paul Debevec that illustrates this process. I’ll post a link if I can find it.
A final plea
Dear Technicolor. Cinestyle is great and opens up our ability to capture higher quality video, but it’s severely limiting if the proper technical details aren’t released to the community to allow us to use it in the correct way. The lack of a clear specification of what the supplied LUT actually does (i.e. the destination colour space) is a surprising omission in this release. Publicly releasing a LUT to go from Cinestyle to Linear would be a simple solution to this problem.
Please, give us more information!