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

[amanieux]

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.

and the cinematic is strange when she walks as if there were some frames dropped

[Payne]

Andrew: 

 

Since a 600D version is in progress do you think a 60D version will come to pass realistically? I cannot afford to upgrade for many years due to more important expenses in life (Wife and kids!). The 60D is however almost the same camera on the inside if I'm not mistaken. Both use a single Digic 4, both write to SD. What do you think the possibility is of a 60D w/ the fastest SD card available outputting to something over 480p in raw?

[JohnBarlow]

What is really interesting to me is that the SD card controller is the bottleneck for the older cams like 550D/600D or T2i/T3i . The bus speeds on those cams are obviously up to it by the fact that they can display images on their viewfinders.

 

I wonder how difficult it would be for a dyed in the wool hardware engineer to sweat on a data bridge to a fast CF card as a hardware mod. There are some clever people in those Arduino type forums who would get this sorted over breakfast.

 

The price of those older cams is now in closeout so even if the mod cost $500 it would be a steal.

[ScreensPro]

A foggy day. Try it with sun beams shooting through the forest. Then you'd wish for the extra nearly three stops at ISO100 of the Exmor sensors. Sure your scene has a decent DR range but it's not crazy high. I should also say that anything downscaled to like 1/8th of a MP look good too.

 

It wasn't fog!

 

That is a forest on a mountian, literally about 100ft from being above cloud level, that is cloud running through the trees... and the light was very, very bright.

 

Maybe not the best example, after all.... but it was very harsh shooting conditions. I'm sure a Nikon would get a bit more highlight and shadow, but what i was saying was that the Canon sensor is enough for most any shot. Certainly up there with RED MX sensor.

 

But all the latest videos are showing just how good these sensors are. Literally can't wait to get my hands on this next ML.

[Sean Cunningham]

 

 

 

Canon looks better... :blink: no moire.

 

Based on the stills I like the color better on the BMCC.  Without processing the only shot I saw that was immediately more pleasing was the road with all the high frequency gravel (but you can't use either of these cameras without processing so what they're showing is a worst case scenario, representing a scene produced by someone who doesn't know what they're doing).  I've dealt with that kind of moire on the 7D and it's no big deal, while still retaining full luminance resolution.

 

The flesh tones on the instructor in the dojo were better on the BMCC.  In fact, even though he's smaller in frame, in the BMCC example, more flesh detail, not just tonality, is visible in and near the highlight area than in the 5D footage.  They even conclude this.

 

Here's the catch with the moire though: you only have to effectively filter color and because the BMCC isn't recording sub-sampled color but a full sample, once filtered you're not far from the 5DmkIII's color fidelity recorded with the hack.

 

edit: but please, don't get me wrong, I'm not implying this isn't great or exciting information.  It's fantastic for people who already own a 5D mkIII (and potentially most other model owners).  I reject the idea that it represents the best choice available for new camera purchase decisions.  Nothing to date suggests that's a rational conclusion.  There are no deal breakers or makers here apart from fixation on an individual aspect while ignoring a host of caveats.

[HurtinMinorKey]

In reference to :vimeo.com/66170436

 

They crushed the blacks so it's not really a really a valid comparison of DR. The BMCC video has been shown to have more DR that 5D3 stills, so I'm not sure why they think the new 5D3 video hack would increase the 5D3 DR beyond its stills capabilities.

 

(looking at the downloaded version)

Consider: @ 1:15 (BMCC) you can see more detail in the wood background, and the highlights are not blown out on his face@ 1:23 (5D3) you can see slightly less detail in the wood and his forehead is over exposed. Looks like a clear DR advantage for the BMCC

[sandro]

I wonder what Canon is thinking about this. Their crippling has been converted into a better version of the high end video cameras they want to protect! AHAHA!

 

Anyway how could this ever work with SD cards? Maybe the 550d is out of the game :( 600d only does 1280x340 with problems.

[hmcindie]

 

Consider: @ 1:15 (BMCC) you can see more detail in the wood background, and the highlights are not blown out on his face@ 1:23 (5D3) you can see slightly less detail in the wood and his forehead is over exposed. Looks like a clear DR advantage for the BMCC

 

The problem in doing those kinds of comparisons is A) you're not looking at the RAW-material and B ) 5d actually has ISO with analogue gain. BMC doesn't. So when you change BMC from ISO 800 you will always lose DR. 5d keeps it A LOT better when going high and also at ISO 100. and C) The exposure point maybe different for the sensors so one of the cams might hold more highlights/shadows at different points than the other.

 

Also when you guys say that BMC "has been shown" to hold more DR than the 5d in stills...where has this been actually shown? DXOmark has only measured the 5d, not the BMC and you can't compare other measurements as everyone does it differently. Provideocoalition measured the FS700 at 14 stops...Is it comparable to the 5d at 11.7 DXOMark stops? No. Way.

[Sean Cunningham]

 

(looking at the downloaded version)

Consider: @ 1:15 (BMCC) you can see more detail in the wood background, and the highlights are not blown out on his face@ 1:23 (5D3) you can see slightly less detail in the wood and his forehead is over exposed. Looks like a clear DR advantage for the BMCC

 

 

Comparing both DNG files I can confirm there is substantially more detail in the BMCC image.  Opening up the stills on their page masks the difference, being a scaled down file.  Viewed 1:1 the 5D image isn't comparable at all looking at the grass.  It's like looking at a stock GH2 compared to something like IV2 or Moon footage.  The blades are green blobs and the detail in the back bushes is entirely impressionistic. 

 

That's looking at the 5D footage as a 1920x1080 image on top of the 2400x1350 BMCC image.  The detail comparison is made worse if you scale down the BMCC footage to 1080P and A:B those two images.  You can still see individual blades of grass where the 5D is sorta blobby.

 

 

The image from the 5D isn't bad though.  I only make it sound harsh because I'm looking at one as a partial overlay on the other.  If you were to see the 5D footage without any comparison I'm sure you wouldn't be missing the extra detail because it's still a pleasing image.

 

They likely picked the single worst scenario for moire that one could, however.  That road.  I can get it reduced with a non-aggressive setting, which likewise eliminates it from all the leaf highlights in the trees, etc., but if that shot were for a film an extra level of attention would be given to it through a mask.

[sandro]

Andrew about your footage...what was the sizxe of the overall RAW files? I read about 7MB per frame for  1920x1080 RAW this would equal to 10GB per minute :|

[Oliver Daniel]

Exciting news for videographers, however my opinion is that a lot of us are going nuts for it far too early. I cant knock the work Magic Lantern do (very impressive) but all this talk of the Canon 5d Mk III being the new 'king of the game' seems a bit silly. 

 

My understanding is that there could be a very hungry RAW recording format working in the Mk III flawlessly. Its super cool because of its low light and full frame sensor, plus the loved up DSLR form factor. If it takes off then its awesome for those who have the time, resources and money to handle the huge lovely files and trust their camera won't melt into oblivion. 

 

If anything, this advancement will become overshadowed by other camera developments very quickly. At the moment, we have a Black Magic Pocket Camera that will shoot RAW and 1080p ProRes out of the box with no potential issues for very little money. If the ML development really does cause a disturbance, I would be least surprised if a company like Black Magic announced another cinema camera with higher frame rates, a better sensor etc.... then suddenly BANG. Everyone wants that!

 

My point is, its a great development. Will i be buying into it? No - because I expect this will be history once an inevitable, 'proper' out-of-the-box advancement is available on the market with the benefits of the Mk III features, and more, with the peace of mind it won't break. 

 

I hope everybody does go nuts for this Mk III RAW thing. Just means we get better cameras in future. :)

[Andrew Reid]

Andrew about your footage...what was the sizxe of the overall RAW files? I read about 7MB per frame for  1920x1080 RAW this would equal to 10GB per minute :|

 

No 7MB is for 3.5K raw frames. It is 4MB per frame for 1080p. Less than Blackmagic.

 

Situation with storage and media is similar to Blackmagic. Just get as much camera media as you can afford, then compress the raw in post to something more management like ProRes or CineForm whilst maintaining the huge image quality leap from straight out of camera H.264.

 

You've seen how good it looks to 24Mbit compressed H.264 from the Vimeo download! Still a big leap from Canon's video image quality even if you throw away the raw files.

[Germy1979]

It wasn't fog!

 

That is a forest on a mountian, literally about 100ft from being above cloud level, that is cloud running through the trees... and the light was very, very bright.

 

Maybe not the best example, after all.... but it was very harsh shooting conditions. I'm sure a Nikon would get a bit more highlight and shadow, but what i was saying was that the Canon sensor is enough for most any shot. Certainly up there with RED MX sensor.

 

But all the latest videos are showing just how good these sensors are. Literally can't wait to get my hands on this next ML.

That shot is still pretty relevant.  You'd lose the 3 row of trees to a supernova of blown highlights if you'd exposed for the stumps on the regular 5D codec, and everything on the lower half of the screen vice versa.

[nathanleebush]

Your final words sum it up best. It's a shame that a company so disdainful of its own customers gets a lucky break TWICE (first by slapping on video mode to 5D II and reaping enormous, unexpected rewards and now this incredible uncrippling by a open source third party which represents all the opposite values as the company itself).

 

I find them kind of despicable as a company for their well-documented feature-damping to protect their premium product lines, but I guess I have no choice but to shell out some money, as the early side by sides with BMCC show a higher dynamic range, better high ISO performance and less color noise.

 

Blackmagic deserves their success for giving the customer what they want and shaking up the entire industry, and I hope they can still reap benefits. I'll keep my pocket camera pre-order, but this pretty much nixes the BM Production Camera for me.

 

I don't really need 4K, and global shutter would be nice, but not essential, and the fact that this doubles as a highly capable stills camera (though I slightly prefer Nikon for stills, for their sensor tech and ergonomics), puts the icing on the cake.

[Andrew Reid]

Comparing both DNG files I can confirm there is substantially more detail in the BMCC image.  Opening up the stills on their page masks the difference, being a scaled down file.  Viewed 1:1 the 5D image isn't comparable at all looking at the grass.

 

That's looking at the 5D footage as a 1920x1080 image on top of the 2400x1350 BMCC image.  The detail comparison is made worse if you scale down the BMCC footage to 1080P and A:B those two images.  You can still see individual blades of grass where the 5D is sorta blobby.

 

I think that's understating the 5D's image. BMCC is still wonderful but it has a serious competitor here.

 

Some of the extra perceived detail in the BMCC's 2.5K raw comes from false detail. The 1080p on the 5D is much cleaner, and less noisy.

 

A good comparison would be to upscale the 5D's raw to 2.5K size and compare on a 2.5K display. I expect a bit more sharpness from the BMCC but is it worth it when the trade off is some false detail and more noise?

 

2.5K resolution is great and 5D3's full frame likely to top out at 1080p, but the 1:1 crop mode could yet reach it's full potential and give us nearly 4K from a Super 35mm sized area with a very fast card. The 5D3's DMA memory in the camera does 700MB/s! That is faster than any SSD and not a bottleneck.

 

I have the BMCC 2.5K and 5D3 with me now and will do some comparisons. Feel free guys to let me know your ideas for tests.

[Sfernald]

I would love to see what dark nighttime or poorly lit shots look like in raw. That is the Mark3's specialty after all. Would be interesting to see how good a poorly lit scene looks.

[mtheory]

 

I wonder what Canon is thinking about this. Their crippling has been converted into a better version of the high end video cameras they want to protect! AHAHA!

 

Anyway how could this ever work with SD cards? Maybe the 550d is out of the game :( 600d only does 1280x340 with problems.

 

 

They're probably trying to figure out how to cripple 7D Mark II which has 1DX hardware components. 

 

 

I have the BMCC 2.5K and 5D3 with me now and will do some comparisons. Feel free guys to let me know your ideas for tests.

 

 

A greenscreen showdown between BMCC and 5D3 would be awesome.

[Marc]

Great news...but...Ummm...does anyone know where you can download 5DMK3 firmware 1.1.3 ? I've just upgraded and can't find the old firmware. I take it the hack will not work with 1.2.1? Anyone?

 

Canon has pulled it from all the sites I've visited.

 

Cheers!

 

 

 

 

 

 

 

[HurtinMinorKey]

 

I have the BMCC 2.5K and 5D3 with me now and will do some comparisons. Feel free guys to let me know your ideas for tests.

 

Shoot in bright sunlight, with bright spots and shadows in focus. expose for the brightest point, then squash the respective dynamic ranges down in post to what is visible in 8 bit. 

 

Does that make sense?

[Julian]

I have the BMCC 2.5K and 5D3 with me now and will do some comparisons. Feel free guys to let me know your ideas for tests.

 

Would like to see some high iso tests from 5D3 in raw. Not 25.600, more something you'd actually use, like 6400.

 

Noise reduction in Lightroom/Adobe Camera Raw works good for stills (I usually only apply colour noise reduction, to keep the grain), does it work well for video or are the patterns not random enough?

 

I'd also like to hear your thoughts on Lossy compressed DNG (see here) converting them is a fast process and saves a lot of space when offloading to your computer. I'm curious what kind of compression you get on average (depends on the detail in the scene), how does it influence post and what about quality? To me it looks visually lossless. I think this could make the whole storage/processing part quite a bit easier.