We had a chance to play with the C300 a bit last week. Before that, I’ve read some blogs and forum posts where they talked about this camera, but it always ended up with commenters bashing each other with numbers, or judging the camera by watching a highly compressed video on vimeo. I’ll try to solve the second problem by posting original files, with all the important details of the given shot written on the attached thumbnail (lens info, ISO setting, f-stop, shutter speed, WB setting, and if the built-in ND has been used, or not).
This was the easier part.
The harder part was: what to do with the bare numbers? Everybody seems to recognize that in video, the higher the number, the better (8-bit vs 10-bit, 2K vs 4K), which is absolutely true in all cases. But my questions are:
- what do these numbers really mean?
- what do they do to my footage, which part of the image do they affect?
- Do I - or the production - really need to, or want to pay the price for a better codec, bit depth, or chroma subsampling? (And price doesn’t simply means money spent on equipment, storage, processing power, but also more time to transcode, edit, grade, export, and so on)
Just for an example: a lot of people wanted RAW, uncompressed video from the Canon 5D mark III. No comment.
Do they know, that this means 207 MBytes (not megabits) of data every second? Do they know that one hour footage would be 747 Gigabytes?
I don’t think so, maybe 10% of the current 5DmkII users would want that. What the others would need - not just the current 5D users - is a perfect 1080p camera, which is balanced between quality, and costs.
And this is where the C300 comes in the picture.
The C300 has a Super35 size sensor, measuring 24.6mm horizontally, and 13.8mm vertically. This means the DOF characteristics will be similar to APS-C sensor cameras, or 35mm film.
When Canon announced that this is indeed a 4K resolution sensor, everybody thought that it will record 4K, but the truth is that it “only” records 1080p. The active pixels during recording is 3840×2160, you don’t need to be a math genius to notice that they simply duplicated the standard 1080p resolution horizontally, and vertically - this is why it’s sometimes referred as QuadHD. This sounds great, but what do we gain from having a higher resolution sensor, if it can only record 1920×1080 internally, and can only output 1920×1080 through HD-SDI or HDMI?
Because 1080p doesn’t equal 1080p.
Again. 1080p doesn’t equal 1080p.
I’ll mention the advantages of the 4K sensor in a 1080p camera below.
If we have a Bayer-sensor camera - most of the current motion and still camera do - at some part, you will de-bayer your raw image data. This is when you lose some resolution, like it or not. This is because these sensors have half the amount of red and blue photodiodes, as green ones (four photodiodes give out one RGB pixel).
You will see, that the second RGB pixel uses the same red photodiode as the first RGB pixel, and the third RGB pixel uses the same blue as the second RGB pixel, and this is when we lose some resolution:
When using Bayer-sensor cameras, source resolution is important, even if post production will happen at lower resolution - let’s stick to the C300 example: even if you will have your film finished in 1080p you will gain advantage of the 4K sensor.
What the C300 does is simply reading four times more pixels - it doesn’t skip pixels as HDSLRs do, it reads all of them -, and this way you will have a clean 1080p image. It is much like recording a 4K file, then resizing it to 1080p on your computer, but all of this happens in-camera without any fuss.
This happens without de-bayering the raw data. It has enough red, and blue photodiodes to get all the data needed for a full 1080p picture, and it has twice the green photodiodes. It’s hard to explain how this happens, but I’m sure it’s a far simpler process, than de-bayering the image.
creating the 1080p image from the QuadHD sensor, without de-bayering
We didn’t have time to make a specific moiré test, but in our footages we didn’t saw moiré problems.
Be sure to check this video from Texas Media Systems for a very good moiré test:
Each digital sensor has a parameter called “dynamic range”, this is the range between the low (black), and the high (highlights) values it can detect. Canon claims that the C300 has 12 stops of dynamic range overall, but again, this is just a bare number. To get an overall image of the measured dynamic ranges of current cameras, checkout Zacuto’s chart
Shadows = the darker part of the image (middle-point is 18% grey)
Highlights = the brighter part of the image (middle-point is 18% grey)
Sensitivity = the ISO setting in this case
Signal/Noise ratio = the difference of a true signal (the picture you look at), and the noise that gets on the footage in-camera
As you can see, dynamic range is always 12 stops, no matter what your ISO is set to. ISO 850 is called “base sensitivity”, this is where you will have the cleanest image overall. With higher ISO settings, you will have a noisier image - everybody who uses a digital camera will be familiar with this -, and with lower ISO settings you will lose highlight headroom, but gain shadow headroom. Given that the camera’s base sensitivity is relatively high - if you’re coming from the still photographer world, it’s very high compared to ISO 100 -, even at dark you can use the base sensitivity, producing the lowest noise possible.
The C300 records super white levels too (values that are higher than 100 IRE as you can see on the waveform).
This means, if you open your footage and you have some blown-out parts,
you can still get some details back by lowering the super white parts under 100 IRE.
An original C-log picture without any correction - if you think this image lacks contrast, you can ask yourself
why C-log is good for me if all I can see is a picture with muddy blacks?
Storing image data in log format comes from storing scanned film in digital format. When you scan a still frame, you want to make sure that every detail that had been exposed on the negative is stored in the digital file too - you don’t want to scan again just because you need to bring back some detail in the crushed blacks. The same applies to recording in log in digital cameras, you don’t want to reshoot something just to have more detail in highlights, or blacks.
Recording in log format ensures that everything the sensor is capable of (the dynamic range we mentioned), and everything you exposed correctly will be stored in your file and will be available to post-production. No detail is lost because some nasty high-contrast gamma curve. Of course you can still under- or overexpose, but all the correctly exposed details will be stored.
Original C-log image, waveform, and histogram
Same image, with a quick contrast adjustment
As you can see, even in a sunny day scenario, the C-log image doesn’t clip the blacks or the highlights,
so you will have the ability to grade it to your liking. You couldn’t do it backwards.
The C300 compresses with MPEG-2 50Mbit/s CBR
and puts it in an MXF
What this means to the user:
Mpeg-2 is far superior compared to h264 compressing very small details like grain-like sensor noise.
source was ProRes422, 400% magnification - check the grain in the pictures
encoded to h264 (VBR 40Mbit)
encoded to Mpeg2 (50Mbps CBR)
50Mbps is higher than what other cameras have in the price range (AF100: 25Mbit, Sony F3 35Mbit), but much less than the ~264Mbit ProRes files coming out of the Alexa. But again: these are just bare numbers, you will have to choose the best compression for your project by testing them.
CBR ensures, that the camera doesn’t go and decide which frame is important and which isn’t, every frame is “important” to the compression.
is the file format it uses, it contains timecode and you get some additional files where the metadata is stored. You can edit these natively in Premiere, and “AMA” it to Avid Media Composer too. I’m not an FCP X user, but as far as I know, they will need to rewrap, re-encode these files. Abelcine blog has a nice article
showing the best way to import the C300 MXFs in to your NLE.
Metadata includes lens data, ISO, shutter speed, aperture, picture style settings and a lot of other details. Right now the only application I could find that reads out these informations, is the Canon XF Utility
One additional side-note is that the files are 8bit internally, and the outputs are 8bit too. This tells us how many luma and color values can it store. For example if we record 10bit files, this means four-times more possible luma, and chroma values compared to the 8-bit version of the same file. You will miss those extra two bits mostly in gradients but to see these, you will need display, and a system that can display 10bit files.
Before I understood what chroma subsampling means, and how it affects the picture, I only knew that 4:4:4 is the best and some pay a lot
for this extra. Actually it is a type of compression too. Most video files store images in Y’CbCr
format, and not in RGB - although these can be converted into each other. Y’ stands for the luma (think of a black and white picture), and the Cb, and Cr carries the chroma/color information. Because the human eye is much more sensitive to luminance differences, it is understandable that if we want to compress something, it should be the chroma information, not the luma.
So the uncompressed chroma data is called 4:4:4 (Arri Alexa internal recording, Sony F3 dual link HD-SDI output with a pricey
external recorder), the next step is 4:2:2 (C300, Canon XF305 internal recording, Panasonic AF100 output), and there’s 4:2:0 (Sony F3 internal recording, Panasonic AF100 recording internally, Canon 5DmkII video).
Put simply, imagine 8 pixels in two rows.
4:4:4 means that 8 pixels will carry chroma information.
With 4:2:2 only four pixels will carry chroma information, other pixels will be averaged.
Using 4:2:0 only two pixels will carry chroma information, other pixels will be averaged.
Well you could say, that’s a lot of compression, but hey, do you see any difference in the next two pictures?
(click to open them full-size)
Ok, well let’s see some details when zoomed in 1600% (check the edges of the teeth, where you will see the averaged pixels)
(click to open it full-size)
Still, very little difference at the edge of the skin. Then why should I even care?
Because even if you don’t see it, this chroma information loss will make your life harder when you try to do chroma demanding tasks.
Let’s bring the images in Davinci Resolve, and try to do a fast key on the skin.
(click to open it full-size)
As you can see, 4:4:4 will result a much smoother, cleaner key. Also these are uncompressed stills (not from the C300), so 4:2:2 or 4:2:0 chroma subsampling with a nasty compression can kill your skin, or chroma-key completely. Altough 4:2:2 is not the best option you can have in the world, it’s still a nice thing to be able to record to 4:2:2 in-camera, without an external recorder.
Of course the 4K sensor also gives you less noise when it’s scaled down to 1080p. But not just the low noise level is excellent, but the noise structure is also different from what we usually see from cameras. Vincent Laforet has a blogpost
about what he calls “organic noise”, be sure to check it out, he loves this.
I created this montage from seven different files, all shot with C-log, without LUT.
Besides the ISO setting I only changed the shutter speed to have the same exposure without affecting the depth of field
One thing I didn’t saw while recording: the purple tint on the right was really purple/blue, check the low ISO original file.
Another example of the noise level in daylight, original pictures without LUT:
And another sample in tungsten light:
ISO 850 contrast adjusted
ISO 12800 contrast adjusted
Notice that even though noise level isn’t increased significally, the second picture of the dirty shoe has a magenta tint which can be tricky when mixing shots with highly different ISO settings.
But we also have a good news: up to ISO 6400 I was unable to see any color difference between the shots.
For judging the noise level be sure to go to the download page
, and check the night shots. This way you will be able to judge the noise level, and the movement of the noise also.
My conclusion that if you don’t have large gradients, or completely black parts in your image, you are safe until ISO 5000. Which is awesome.
But even going beyond ISO 6400, you will have a unique noise, I think some will even turn up the ISO on purpose for this.
There is no universal truth when it comes to camera tests: what’s best for one can be the worst nightmare for another. And of course each company has it’s own fanboys who aren’t even looking at any footages, and judge by meaningless numbers. Even though there are still one thing I can say for sure: noise is exceptional not just in it’s price range, but of any current camera. And again, not just the noise level, but the structure of the noise. If you see some noise at ISO5000, than you recorded a completely black night sky and looking at the ungraded C-log footage without LUT.
As for the other features:
Of course you will be able to find a camera with a greater dynamic range (Alexa)
Of course you will be able to find a camera that shoots 5K resolution at 96fps (Epic)
Of course you will be able to find a camera with 10-bit, 4:4:4 outputs (Sony F3 with the S-log package, and with an external recorder)
But if you don’t have endless sources of money, time, and manpower, and you want to choose the best camera for your jobs/projects, then you won’t choose the mathematically best camera. If you’re looking at the whole picture - not just the camera itself, but the prices of the accessories, media, lenses, and also check the prices of storing, processing the recorded footage - , you will choose the camera that fits in your workflow perfectly, and gets the job done as good as you can.
For a very similar scenario but in a completely different world, check this Top Gear episode
to see where does it get you if you take the mathematically best option without considering the whole picture.
James May brings a race car for a picnic…
Back to the review, I think a lot of people will choose the C300 because of the high-ISO performance, the in-camera compression, the in-camera chroma subsampling, trustability, easy handling, the cheap media to record, the ability to record on two cards simultaneously, and the price. All I can say is that as soon as you can check it out in person and see it yourself.
ps: if you do have endless sources of money, please contact me A.S.A.P.:)
Thanks for reading!
if you would like to contact me, my email is in the footer
If you would like to use any image, animation, footage, or the text itself, please contact me