Messages

Hmmm, it seems that when I copy %1 newest build over a freshly installed unified ML build from June 19th, the camera freezes when raw_rec is being loaded. Do I do something wrong here?

This lowers CPU by like 10% so should help with that too. Yes, I kept the restriction on frame sizes.. easiest thing would be just to add some back and try it out. Anyone really want the couple of pixels?

In 1x mode, yes please, since even with the May 28th build, it has never been a problem to record the full sensor size of 1584x1062. I think the 50 pixels matter.

...Having said all of that, remember that each camera features an ability to shoot 1:1. At that point, every camera records with a pixel to pixel perfectly sharp image. I suppose the quality of the 50D and 5D II are similar to the upcoming Blackmagic Pocket Cinema Camera when shooting in a 1:1 crop mode. However, the BMPC camera will have a much newer sensor and will feature better high-ISO noise performance. Then again, it's 12-bit and not 14-bit video. Technically it may be possible to apply noise reduction to the Canon 50D and 5D II and have the same low-noise performance based on the extra information.

What I really don't like about the BMPC is that you are stuck with a 3x sensor crop. It severely limits your choices. On a 50D, you can choose between shooting wider at a slightly reduced resolution, or 1:1 with a 3.975 sensor crop. I also really like the fact that on the 50D, you are not limited to 1.78:1 aspect ratio.

It also needs to be seen if Black Magic will actually manage to get full 12-bit raw recording working on SD Cards. Currently, the BMPC "only" shoots 10-bit ProRes 4:2:2.

From a price point, I think you either go with a 50D, and shoot good quality RAW at a VERY LOW price, or get a BMCC and have real 2.5K resolution for an affordable price. Getting an BMPC suddenly doesn't make much sense to me anymore.

Hey Crash-Film... Got a response to your card write-speed issue.

The math doesn't seem to add up when it comes time to how a card performs... Put a Komputerbay 64GB 1000x in a Canon 5DIII and experience 125MB/s write speeds. Place the same card in the 50D and experience approximately 70MB/s. The difference can be in the EXFAT format and the FAT32 format limitations.

The biggest limitation on the 5DMKII and 50D is not file system related, but the memory bandwidth of the internal RAM. If you benchmark the write speed of the 50D without LiveView activated, it easily surpasses 90 MB/s on my 1000x KomputerBay 32GB Card. Once you activate LiveView, sensor data gets written into SDRAM, and the Compact Flash DMA has to share the memory bandwidth with the sensor image DMA stream, dropping to about 67 - 70MB/s.

The 5DMKIII has a revised DIGIC with faster SDRAM memory access, which is the reason why it can still write >90MB/s with Live View on.

is 1920x818 (2.35:1) also in 1X mode? You mean 1:1 mode right..

1920x818 in 5x zoom (aka 1:1) mode, full sensor area (1584x1056) in 1x zoom, (3:1) mode.

Are you serious 1920x872 before it was 1584x720 that's great news is it stable?  Where did you find that build because @GregoryOfManhattan only posted the one from June 9th please send direct link wanna give it a run..  I love that aspect ratio.

With June 9th build, I could "only" do 1920x818 (2.35:1). 1920x872 is with the current June 11th build. In 1x mode, I could easily do 1582x1056 with all builds so far.

Just grab it on the first page of this thread.

[BMCC]

The 50D is taking on Godzila scales. Canon is so freaken lucky to have made this little monster. Otherwise we'd all have BMCCs . . .

I have to admit that the BMCC still has some serious advantages over the 50D (2.5k, SSD, Audio).

However, I see very little to no reason now, if you want to shoot RAW on a low budget, to buy the BMPCC. Even less every step we get closer to Full HD.

[KomputerBay 32GB]

Just tested the June 11th update with my KomputerBay 32GB card.

In 5x Zoom mode, 50D maxes now out at 67.7 MB/s.

On the other hand, 2.20:1 aspect ratio in full HD (1920x872) has now become possible with the 50D. Very nice!

.

[ADDITIONAL]

...and that equals to x / 16, and likewise, changing to 12bit means x/4

Or in other words: shift right by 4 or 2 bits.... 

Concerning precalculated values: if CPU bus speed to memory is an issue (just look at the plain memcpy values), then I don't see how ADDITIONAL external memory fetches from a LUT will help this cause in any way.

I also see no way how 12-bit/10-bit compression will ever be achieved with the CPU. It was only designed to do GUI/filesystem management related stuff. The real processing is done by highly dedicated signal processing blocks inside the DIGIC. As said before, the only chance to get this working is finding an undocumented register bit in DIGIC which reduces the bit width of written sensor data.

With no documentation on DIGIC, this is like finding a small coin inside a barn full of hay, without any kind of guaranty IF this coin exists at all.

[Canon EOS 50D]

Maximum speeds Canon EOS 50D (KomputerBay 32GB card):

1x mode: 1584x1056 (66.9 MB/s) -> ?? MB/s achievable
5x/10x mode: 1920x818 (62.9 MB/s) -> 65.5 MB/s achievable, limited by DMA

This is with June 8th build of ML.

Another update: when the KomputerBay 32GB card is warm, I get a constant 65.5 MB/s write speed in 5x zoom mode, allowing to shoot about 1450 frames in 1920x876 mode.

In standard 1x zoom mode, 1584x1056 at 66.9 MB/s is no problem.

So, as I thought, at this point, the limiting factor in the 50D seems not to be the CF controller, or the card, but the DMA bandwidth being left available in Live View 5x / 10x zoom mode. No matter what resolution you choose for recording, DIGIC4 always seems to write 2000x1080 pixels to SDRAM in 5x mode, which puts a heavier burden on DMA bandwidth than 1x mode in 1584x1056.

If one could alter that value to match the recording size, I think recording between 1920x872 and 1920x960 continuously might become possible with the 50D.

But, as it stands now, being able to record RAW, uncompressed video in 1920x818 on a 5 year old DSLR ist just plain amazing!

Raw zebras are CPU intensive. Before you had YUV zebras with zero cpu. Just turn them off while recording.

Thanks for the clarification! Do you think it would be worthwile to make this option switchable in RAW recording mode, or are the YUV zebras too imprecise for any useful application in RAW mode? I personally found them useful to make quick changes to my exposure settings when the lighting conditions changed constantly.

EDIT: AARGH, I'm dumb, just found out how to switch to YUV Zebras... 

Btw, I also had time now to test the June 8th build.

Compared with May 28th build, I now have to switch off the RAW zebras and the colour histogram to yield a comparable performance. My KomputerBay 32GB card maxes out at 62-63MB/s in 5x zoom mode, 1920x818.

On the other hand, the framing and recording issues in 5x zoom mode are gone now.

It seems also that initially, the write performance is way lower, up until you run the "Card Buffer Benchmark" test in the Debug menu for about 1-2 mins, with LiveView turned on. It seems that ML picks then the best buffer size for your card.

I believe its a bit over 3x. Could be mistaken. That puts it in the BMPCC territory.

Using 1920x1080 pixels in 5x mode would yield a crop factor of about 3.975. The sensor area being used is a little bit smaller as a regular 16mm film frame. Already did some calculations concerning this earlier in this thread.

I think the 50D is far more flexible now than the BMPCC. I'm so glad that these developments here happened before buying one.

Hello guys,

so, I tried shooting some test footage in 2.35:1, 5x mode (1920x818) today with May 28th build, but there still seem to be some bugs left. I apologise in advance in case these issues have been resolved in the meantime.

Firstly, the framing is not correct. Instead of being in the middle of the sensor, it seems that the image is captured at sensor pixel coordinates X=Y=0.

Another problem is that, although I constantly recorded in 5x mode, the written RAW data is sometimes being interpreted as 1592x678, but the image itself seems to be still 1920x818, with a wrong stride being applied in the raw file.

Here is an example DNG of the wrong image data:

http://www.filedropper.com/m000000

As a reference, here is a DNG with correct image data (roughly the same scene):

http://www.filedropper.com/m000000_1

I think it has something to do with the fact that I constantly switched between 5x and 10x mode to check the focus. But I always started recording when the Live View was in 5x mode.

[optimize for -> exact fps]

Turn optimize for -> exact fps on in FPS Override Menu.

[KomputerBay 32GB card:]

Just a few more tests.

KomputerBay 32GB card:

Full sensor in cropped mode (1592x1062): no frame drops. Excellent to choose framing at a later point, like Super35 mm.
5x zoom mode in 2.35:1 aspect ratio (1920x818) : no frame drops. Very nice, full HD shooting in Cinemascope possible!
5x zoom mode in 2.20:1 aspect ratio (1920x872): frame dropping. No 70mm aspect ratio available... 

[turn on hacked mode!]

Finally, my KomputerBay 32GB card arrived! 

Just shot a few tests @1592x896. Buffer always stayed at one star, no big deal, camera and CF card bored as hell!

But important: turn on hacked mode! Otherwise, it won't work!

More experiments following...

Also tried it (with a POS 512MB Kingston CF card), no success! 

Thank you very much! Seems that Pelican's guide many people are using as a reference for Windows compilation is out of date.

Just a quick question:

is it safe to compile the latest contributions with GCC 4.7.2 from the Windows Yagarto Toolchain? It seems that currently, 4.7.3 is used, but didn't find a precompiled build for Windows (yeah, I know I'm lazy...  )

Thanks for your help!

My latest test, guys.
This shows off the latest 50D build with the raw, untouched images and fully graded images. Also a small comparison between Canon h.264 video and RAW image frames.

https://vimeo.com/67415324

BTW, I'm in to donate as well!

I really like that grainy 16mm look. And detail is quite good, leaps and bounds above the H264 video image. I think that from a price<>image quality point, the 50D is by far the best alternative, and a serious threat to the upcoming BMPCC.

And since 1% said that there likely won't be a way to increase it, I would simply stop dreaming and be happy with what we already have

Only unhappy guys have the ambition to push boundaries. 

When my KomputerBay card arrives hopefully tomorrow, I'm all set for testing and hacking.

[3,975]

Crop factor in 1920x1080 should be around (4770/1920)*1,6 = 3,975. So a 12mm lens would be equal to 48mm.

On the other hand, it would enable to use regular 16mm /Super 16mm film lenses for filming, because the size of the used sensor area would measure around 9,05mm x 5,08mm, which fits into the 16mm size boundaries.

So, in theory, IF the 50D code could be optimized to write 1920x1080 24fps in cropped mode, it would be a perfectly useable digital replacement for a 16mm film camera. Even better, you could switch between full resolution in 16mm mode, or slightly lower resolution for shots which require shallow DOF.