BIG NEWS - Hands on with CONTINUOUS raw recording on Canon 5D Mark III

[Lamartine Cabral]

Ok, so there are different explanations as to what is happening from the sensor to what is recorded.

I will use my limited understanding to try and suppose what might be happening.

 

The 5d mk3 is interesting in that its native horizontal photosite resolution is 5760 which happens to be 3 times 1920.

which means that 9 photosites are being used (or should be being used) to get the values for one output pixel.

 

The image bellow shows how this would apply to a small section of the sensor.

 

5d mk3 bayer pattern to 1920.bmp

 

As you can see there are 4 possible pixel types.

 

Thomas could be right in saying that the output is YUV.

However RGB to YUV conversion could happen regardless of how the RGB values are obtained from the sensor in the first place.

 

There are 3 rough methods I can think of for obtaining the RGB values:

 

Method 1.

The most information would be obtained by reading all photosites of a particular colour and averaging them to obtain a colour value for that output pixel.  All photosites would be used.  So for example in the top left group of 9 photosites, 5 green samples would be averaged to for the green value, and 2 blue and 2 red samples would be averaged for their respective values.

 

Method 2.

If however Andrew is correct that no calculation is done on camera,

then this would mean only 1 photosite from each group of 9 would be used for each output colour value.  Meaning you are throwing away 2 thirds of the sensor information and therefore 2 thirds of light gathering ability / noise performance.

 

Method 3.

Or if it turns out the final image still needs debayering then this means only one colour is being recorded for each group of 9 photosites.  Which means only 1 photosite out of 9 is used:      8/9ths of the information thrown away!

 

If Method 3 is the case then I would say this is quite poor.

A 4K raw frame from the 4kBMPC from would offer 4 times the resolution (even after downscaling the BMPC) and 3.1 times the light gathering ability (taking into account sensor area difference of 2.9* but not photosite size)

 

If method 2 is true then the 5D MK3 and 4K BMPC will be fairly evenly matched in terms of resolution and light gathering ability.

 

If the Method 1 true, which is probably unlikely, then the 5D would be simply beastly.

 

It will be interesting to see what the case is.

I think that a way to put this in test is just compare the amount of noise of a 1920x1280 fullframe raw with a 1920x1280 1:1 raw. Can anyone do this?

[Zach]

I can't quite wrap my head around the workflow. Is it spitting out a bunch of still frames, or is it a video file of some sort? 

[Caleb Genheimer]

I'm still holding out for my "perfect cam"

 

4K S35 RAW SSD XLR ND Global Shutter Mirrorless with good dynamic range and preferably with anamorphic-friendly aspect ratios too. The new Blackmagic Production cam comes awful close, except for the nonsensical EF mount and no ND. 

 

I still have hope that it will show up one day. 'Till then, 'tis GH2 for me. I just saw Upstream Color in a pristine digital cinema, and it lacked for nothing in image quality. It's certainly not all about the camera, but the right tool can make the job much easier.

[VanWeddings]

does anyone else see something funky where the woman is walking across the green landscape. her limbs seem to leave some shimmering behind, like a twixtor artifact.

[blackrat]

The new video footage takes it even to another level yet. Simply speechless.

[Husah]

Canon sensor even in RAW Still mode only has 11.7 EV of dynamic range at only ISO 100 (it's less at higher ISOs)., 

 

http://www.dxomark.com/index.php/Cameras/Camera-Sensor-Database/Canon/EOS-5D-Mark-III

 

11.7 is the best case in still mode. In video mode  even in RAW it's most likely going to be lower than 11.7

 

Black Magic Cinerama camera will easily beat it ... 

[hmcindie]

Canon sensor even in RAW Still mode only has 11.7 EV of dynamic range at only ISO 100 (it's less at higher ISOs)., 

 

 

That's not exactly true. There is quite a lot of shadownoise with Canons compared to for example the Nikon D800 but DXOMark just cuts off that shadow and says that's it when in reality there is still dynamics and range there. And in higher ISOs the 5dmarkIII will bypass the D800, that's no small feat. For someone like me who likes to shoot in near darkness, that dynamic range in higher ISO's is a real boon.

 

I really find it odd that people say these sensors are bad. Are you guys actually shooting with them? I have and I also shot with the D800 for awhile. My choice between them would be...the 5dmarkIII. D800 does indeed have a more capable sensor, but a camera is more than that.

[hmcindie]

It's not a better raw and it has, just going on spec sheets, a stop less DR.  It's (5D) 14bit versus 12bit, but it's also 4:2:2 versus 4:4:4. 

 

What are you on about? There is no 4:2:2 and 4:4:4 when going raw. 

[nahua]

I can't quite wrap my head around the workflow. Is it spitting out a bunch of still frames, or is it a video file of some sort? 

It produces a file M000000*.RAW, a single file.  Then you use raw2dng.exe to convert it into still DNG frames, in the hundreds or thousands.  Very crude at this point but it works.  I'm following Andrew and using After Effects to compile them all into a timeline and then rendering into ProRes 422HQ for edit in Premiere Pro.

[nahua]

The sensor isn't doing any additional work.  It's generating this information already.  What's happening here is that work isn't being wasted, disappearing into the ether, it's being diverted to the CF.  It's generating this image and then destroying it, in stock form.

 

The sensor doesn't handle I/O.  I would be really surprised if this isn't actually less work than the heavy lifting of even the crummy MP4 encoding it was doing on the fly, in stock form.  It wouldn't shock me if the camera ran cooler.  This is an I/O bound task more than a processor bound task.

My camera gets warm and the CF card is warm too.  Getting me worried since this is my main camera for stills.  I haven't had it get warm before when recording H.264 files.  I'm trying it out but I'm always checking to make sure it doesn't melt down on me...

[OlegKalyan]

Hi Andrew, could you test Raw 5dmk3 vs C100, in a test 

[MatthewP]

Hi Andrew, could you test Raw 5dmk3 vs C100, in a test https://vimeo.com/66063838

I see lots of false detail, aliasing, Gh3 looks better to me, honestly, cleaner, more detail, 

 

could you film a chart, please? 

Do not mean to be a show stopper, but lets see the facts, not hype!

 

 

It's the way he's managed the footage in post. This is how it should look:

 

 

No aliasing or false detail there!

[Husah]

That's not exactly true. There is quite a lot of shadownoise with Canons compared to for example the Nikon D800 but DXOMark just cuts off that shadow and says that's it when in reality there is still dynamics and range there. 

 

Dxomark is the most credible RAW sensor data on the planet. According to dxomark,  Canon 5D Mark III has only 11.7  stops of DR in still mode at ISO 100. I am not sure how much line skining happens in video  mode, but I am sure some of it does happen. That implies dynamic range far lower than 11.7 for video. 

 

What that means is that you really don't need 14-bit RAW with 5D Mark III. Even 10-bit would be enough to capture all usable data that that sensor can capture. 

 

It's no Alexa 

[riccardocovino]

It's the way he's managed the footage in post. This is how it should look:

 

 

 

No aliasing or false detail there!

 

Even if the result is incredible the aliasing is quite evident in the white wall surrounding the parking lot.

[riccardocovino]

Huge thanks go to A1ex, g3gg0, 1% and the whole Magic Lantern team.

I have no other words to describe it - this is huge news.

Magic Lantern have done the seemingly impossible and given us a continuous raw recording mode on the Canon 5D Mark III. Once activated in the menus the 5D Mark III becomes essentially a full frame Blackmagic Cinema Camera and amazingly mine has not yet exploded. No more short bursts of raw, this is the real thing.

Read the full article here

Around 2:10 the grass detail looks very strange on the top half, I'm wondering which kind of interpolation is computed.

[hmcindie]

Dxomark is the most credible RAW sensor data on the planet. 

 

So why then is the Canon 7d / Sony nex 5n results completely different than what I see on the images? First of all, Canon 7d is pretty much 100% the same image as the Sony nex 5n. Same dynamic range (Sony slightly better in ISO 100, Canon slightly better in ISO 6400), Sony clips highlights a bit worse but holds shadows a bit better, 7d the other way around. I have both of those cams so I know how they look. I consistently get better images with my 7d, even with the sames lenses.

 

So why are the scores so different? The difference is BIGGER in the score than D800 vs 5dmkIII! That doesn't make ANY sense and it invalidates DXOMark. The difference in sensor quality between the D800 and 5d is easy to see. Not so with the 5n and 7d. And it's not the only weird thing going on there. For example, their Olympus results have always been quite odd.

 

Also, FYI, line skipping does not affect DR.

[ScreensPro]

These guys have the DR down at 12.5 stops.

http://www.imaging-resource.com/PRODS/canon-5d-mkiii/canon-5d-mkiiiA5.HTM

 

My eyes tell me this is about right.

 

DXOmark works solely on signal to noise ratio, iirc, which doesn't tell the full story.

 

Regardless,,, anyone who has ever used a Canon for still shots know that the DR is very good and all sorts of things can be boosted and pulled back, in post. Time to stop worrying about numbers and remember we have great, built in testing tools, OUR EYES!

[ScreensPro]

This was on my 5DII.... if that isn't enough DR, i don't know what is. (That is very bright cloud on the way up a mountain, it would ruin the shadows in h264... The tree stumps would be almost black).

 

[OlegKalyan]

could someone who has access to camera and ML patch shoot a chart?

[jgharding]

If this could be made fully reliable it'd make the 5D iii a great camera: you have standard mode for all your events and quick turnaround stuff, and RAW for your big projects. Ace!

 

There does seem to be a lot of odd artifacts at the moment (those vertical lines in out of focus and distant areas, some very jagged edged and square noise) but I'm sure it can be sorted, they've done so much work already.

 

I've used Magic Lantern a ton, donated way back in the day and have been spreading it like wildfire, I'm glad they just don't stop!

 

One request would be: APSC/S35 crop mode. I have some old cinema glass that wasn't designed for 135 frame. Is this possible?