The REAL difference between normal DSLR video and 5D Mark III raw video

[Andrew Reid]

To learn more about using raw video on the 5D Mark III, get your copy of the EOSHD 5D Mark III Raw Shooter's Guide here

What is the real advantage of installing Magic Lantern on your Canon DSLR, specifically in this case the lovely 5D Mark III?

Is it possible to finally SHOW it? Yes it is.

Here is the most in-depth comparison yet between the standard video mode and raw and why the image quality is worth your attention.

[url=http://www.eoshd.com/content/11395/real-difference-normal-dslr-video-5d-mark-iii-raw-video]Read the full article here[/url]

[terozzz]

Ok.. I get it... Good bye my money, welcome RAW.    :/

[andy lee]

the differance is quite remarkable!! very useful article - thanks

[Andrew Reid]

Ok.. I get it... Good bye my money, welcome RAW.    :/

 

Magic Lantern is free!

[terozzz]

Magic Lantern is free!

Well needs to get 7D/5D MK III first :/

But still wonder if Ninja 2 etc Recorder will make my D5200 any better, havent seen any good tests about it..

[Andrew Reid]

Well needs to get 7D/5D MK III first :/

But still wonder if Ninja 2 etc Recorder will make my D5200 any better, havent seen any good tests about it..

 

Uncompressed HDMI + external recorder doesn't make it any better really. Everything on the page I wrote here still applies aside from compression. Even then, uncompressed HDMI doesn't help the compression of colour, down from 14bit to 8bit and 4:4:4 to 4:2:0 or 4:2:2 (I doubt it REALLY is 4:2:2 on a Nikon to be honest). Resolution doesn't improve either as the camera is still doing a bad debayer and running the image through the image processor before it reaches the HDMI port (to apply the crappy picture profiles and noise reduction for example).

 

I actually contest that HDMI is truly 'uncompressed' on any DSLR.

 

It has to go through the image processor and that involves some compression.

 

Maybe on the Sony A7 we will get a bigger benefit, as the image might go from the debayer stage direct to the HDMI port with no other image processing applied… That's what the Sony guy alluded to in my chat in Berlin any way.

 

5D Mark III raw bypasses DIGIC altogether. It's literally a sensor readout in a file. Everything happens in post, including the very nice Adobe Camera Raw debayer. You can throw huge power at the image in post, power the DSLR itself just hasn't got within the DIGIC 5+ CPU.

[tehgeek]

Magic Lantern is free!

 

But the cards to make it usable are far from free.

[Nikkor]

Nice "review", it shows all the benefits, specially the ones people seem to forget like the choppy motion blur which makes you nauseous, this one is a no-go for me with h264, when I look at motion blur in raw videos it looks almost like film to me, the rule about 6 seconds panning from one to another side doesn't apply with raw video (from my little understanding) even if it is taken at 24fps it still lacks that choppyness from h264 video files.

 

ML raw is great, and the extra cards are only a problem if you have to shoot a lot of takes in a short amount of time, but with this at least you have the choice: "almost RED quality" for a fraction but with extra hassle or just go BMPCC for the other situations. (once they release RAW) 

 

One suggestion.Andrew, the color comparsion where the RAW appears all yellow and the h264 looks about right, I would put another example matched to the h264 to show how you have controll and how much better it looks. I guess it's obvious but some people might get confused.

[maxotics]

the differance is quite remarkable!! very useful article - thanks

 

Finally!   :)  If Andrew can convince you he can convince anyone!  (Of course, if I was shooting my niece or nephew's concert I'd rather have a G6).  

 

Another awesome article!  

 

A comment on this: "Both JPEGs and raw video start off as raw image data but because the 5D Mark III does not shoot JPEGs at 24fps, it has time to think. Therefore the debayering quality is far better on a JPEG than it is on a frame of video."

 

In this context, there is probably little difference in bebaying quality between photo and video (though, of course, debayering in post is always preferable) .  A more important difference is the size of each photo/frame saved to JPG.  I'm going to have to work with numbers off the top of my head, but should hold up.  Anyway, a 1920x1080 photo or video image would probably take up at least 1 megabyte of JPG space.  That means 24 megabytes per second, if you're using those images for video.    Even if motion compression can reduce that 50%, you're still dealing with 12MB x 60 or 720 megabytes per minute.   let's just say a gigabyte.  

 

Supposedly, All-I should be able to do a version of this, but the footage I see never looks very good.  I believe the reason is All-I video is like compressing frames in the highest-compression, lowest quality settings.   

 

Even today, most consumer cameras seem to keep at that 28Mbit/sec video rate.  There's just no way to fit photo quality video in at that rate.  Most people, myself included, wonder why they can't come up with say 6MBS, or something like that.  I think the reason is the time it takes to compress.  Too much CPU power is needed.   Anyone who has ever compressed anything knows it takes no time to de-compress, but a lot of time TO compress.

 

The only thing the camera manufacturers can do (to keep the lowest-common denominator of shooters happy), is, as Andrew elegantly puts it, "trashing" the image (throwing out data before compression).

[markm]

Finally!   :)  If Andrew can convince you he can convince anyone!  (Of course, if I was shooting my niece or nephew's concert I'd rather have a G6).

Heh heh.

[alexcosy]

Nice article, as always :) , i guess my question would be: How can i perceive 14 bits depth on my 8 bits monitor. That's something i have a hard time understanding.

[andy lee]

Finally!   :)  If Andrew can convince you he can convince anyone!  (Of course, if I was shooting my niece or nephew's concert I'd rather have a G6).  

I was convinced 4 months ago when the hack first came out - but the workflow is still too time consuming so my MK111 is still a nice stills camera , G6 is the main workhorse

[Andrew Reid]

But the cards to make it usable are far from free.

 

No but last time I checked, the 64GB KomputerBay card cost $120 and the Alexa with Codex cost $80,000. Take your pick!

[Andrew Reid]

I was convinced 4 months ago when the hack first came out - but the workflow is still too time consuming so my MK111 is still a nice stills camera , G6 is the main workhorse

 

I'd go for the G6 over hacked GH2 as well. The codec is actually better implemented on the G6 and you don't get a nasty Russian man with personal issues calling you Voldermort into the deal. Bonus!

[markm]

No but last time I checked, the 64GB KomputerBay card cost $120 and the Alexa with Codex cost $80,000. Take your pick!

Yes but aren't the computer bay cards supposed to be risky?

[maxotics]

Nice article, as always :) , i guess my question would be: How can i perceive 14 bits depth on my 8 bits monitor. That's something i have a hard time understanding.

 

That tripped me up too.  When we speak of 8bits for monitors (255 integer) we mean, per color channel.  As you'll see when you set your display settings, you want 24bit (3 x 8 (byte)).  8bit video delivers about 16million color values--the range of human vision.

 

When your camera takes a photo/video frame, each sensor pixel is taking a 14bit reading, that is 1 to 16,383 (or something like that).Each pixel actually reads only a red, green or blue value.  Another complication.   Anyway, that number is ultimately converted to 0 to 255. So you're giving up a lot of accuracy about just how much color there was.

 

In RAW video, you get those 14bit values BEFORE the cameras converts them into 8bit equivalants.  Complex subject, hope this gets you on right track.

[alexcosy]

That tripped me up too.  When we speak of 8bits for monitors (255 integer) we mean, per color channel.  As you'll see when you set your display settings, you want 24bit (3 x 8 (byte)).  8bit video delivers about 16million color values--the range of human vision.

 

When your camera takes a photo/video frame, each sensor pixel is taking a 14bit reading, that is 1 to 16,383 (or something like that).Each pixel actually reads only a red, green or blue value.  Another complication.   Anyway, that number is ultimately converted to 0 to 255. So you're giving up a lot of accuracy about just how much color there was.

 

In RAW video, you get those 14bit values BEFORE the cameras converts them into 8bit equivalants.  Complex subject, hope this gets you on right track.

 

Hey, many thanks for answering, my question was more something like: As anyone, the actual display i'm looking at right now (99% of us are not on10bits Eizo or FSI displays I guess)  is 8 bits, How can i possibly, on an 8bits display, perceive 14 bits color gradation. I get that the video itself contains the information, but if the display is 8 bits... i only see 8 bits color right ? Just as if it were black and white, i wouldn't see color, no matter what.

I know it's tricky and i guess there is something i'm misunderstanding... That's why i'm asking, so i can understand fully this, and maybe at some point give the answer to someone asking the same thing.

 

 

Or, were you saying that an 8 bits display is not the same thing actually more colors than 8 bits video ? it's alway 8 bits per channel no ?

[MaxAperture Films]

Man, what a difference... and the NR response is phenomenal.  My only hangup is the workflow and data requirements.  How does the raw image fare with a transcode to Cineform... is all the latitude still there in exposure adjustment?

[gloopglop]

seriously one of my favorite pieces yet on EOSHD, thank you Andrew

 

I love how youre able to make nuanced points about subjective aesthetic issues while incorporating the technical details that make it happen

 

all that and you crack me up too

 

HDR puke

 

lmao

[maxotics]

How can i possibly, on an 8bits display, perceive 14 bits color gradation.

 

You can't!  In the way I mean it, 14bits of each primary color.  Looks like I have to go into those "complications".  Camera sensors are monochrome.  They read light be placing little filters over each pixel, either red, green or blue.  Each pixel then "borrows" the 2 other colors it doesn't have.  So if it is a red pixel, it take the green and blue color information from neighboring pixels to create a full 24bit color.  (BTW, they don't work with RGB but YUV, oh this stuff is so f'ing torturous!)  But, for explanation sake...

 

Let's say we're in a perfect world.  You have 3 color values, each from 1 to 16,000 (red, green or blue).  That means, from those, you can create a full color at 16k x 16k x 16k depth, or 4 trillion!  You can't discern 4 trillion colors.   So now you have more color information than you can physically see.  In the end, we always need to reduce to 16 million.

 

Here's the rub.  You can't see 4 trillion colors.  The camera can record the 16 million you can see in 8bit video.  So what's the problem?  The camera may not chose the 16 million color values you would chose from a palette of 4 trillion colors.  As the article shows, it is never smart enough to do that.

 

RAW allows you to  SELECT which colors to scale down to your 16 million painting.  As Andrew said, do you want to start with 4 shades of pink, or 255?  It's all about CHOICE in what you want your final 8bit channel image to be.

 

Are we getting there?