Messages

You'd be better off exporting PNGs than Jpegs.  VirtualDub will use those too, and obviously it's lossless compression vs lossy for Jpegs.

It won't work - ACR can only export TIFF, JPEG, DNG, and PSD. I am using maximum jpeg quality (essentially uncompressed), and I don't see any difference in the video when compared to the one made from PNGs. (I did test another workflow, where I did ACR->TIFF->ImageMagick tools (upscaling + unsharp mask)->PNG->Virtualdub, but it involved more steps, and the ImageMagick step was very slow.)

[set the sharpening to 0]

Great work, thanks! I'll be one of alpha testers for this MLV stuff.

A question: does each frame include things like aperture stop and focal length? This is critical, as that's what lens correction module of ACR needs to get read of those annoying aberrations, distortions and vignetting. (Also, having the lens model would be very helpful; but if not, one could always add it later with exiftools.)

I am trying to come up with a "photographer-friendly" workflow for 50D RAW video PP. Specifically, I want it to be built around Adobe Camera Raw (packaged with PS). It actually works very well, except for one deficiency: my version of ACR (6.x) doesn't provide the upscaling option to get from uncropped video to 1920x1080. (It does provide a few sizes, but no way to customize that - weird!). If not for that, one could have a complete solution in ACR - it'd do all the grading, lens and perspective  corrections, then upscale to 1080p and sharpen the frames. No such luck - the last two steps (upscaling and sharpening) would have to be done elsewhere.

Virtualdub is probably the only video tool I know well (have been using it for ages), so I was trying to use it yesterday to supplement ACR. So the workflow is:

• Convert RAW to DNGs
• Load DNGs into ACR, do the corrections, but set the sharpening to 0 - I think this is important, as any sharpening (even the default one) has to be reserved until the end, after upscaling - am I right here?
• Save the frames in ACR as JPEGs - this seems to be the only way to directly bring them to Virtualdub (it doesn't know TIFFs).
• Open the JPEGs in Virtualdub, load two third party filters - Smart resize (I use exact bicubic resize to 1920x1080) and MSU Smart sharpen (I am debating between "Small overall" and "Overall" levels of sharpening), and then save this as an AVI video with the codec of your choice (I use Xvid for now)

At the end, I do see moire/aliasing artifacts, but they seem to be no worse than in the original size video. No flickering AFAIKT. And apparently no new artifacts due to upscaling - e.g. a high contrast diagonal line doesn't appear jagged. I think that's very crucial.

Does this look reasonable?

I did a bit more testing with my new card (Lexar 32GB 1000x), and the October 11 build. In crop mode, I can reliably go up to 1920x1080x24p resolutions in 16:9 ratio (83.0 MB/s; no dropped frames) and 2000x910x24p in 2.2:1 ratio. At even higher res, 2000x1080x24p, I get ~1000 frames before a frame drop.  I have to turn Global draw off and enable Hacks to make this work. I am happy about this results.

But many points are not clear to me. I don't see crop marks when doing 16:9 cropped video - is it possible to enable them? I think I got it - it needs to be enabled.

I tried the turning Digital dolly on, but it didn't do anything for me in crop mode (I pushed the joystick around) - neither before nor during recording. Is it supposed to be functional?

I was recording first in C1 mode, but then switched to Av to have DoF control. It looks like in Av mode I can't do any changes to aperture or shutter speed during the recording - is it normal? I am looking wright now at the 1920x1080 crop video, and it looks very noisy - does it mean that ML boosted ISO a lot? How to make sure I use only low ISO? (I am using "optimize for low light" RAW video setting.)

Thanks!

You can choose which area is used in crop by moving the box as usual before you record.

There is a digital dolly setting in the raw menu, which is supposed to let you use the joystick to pan the sampled area during cropped recording.  In my case it doesn't work as expected.  Once the option is enabled, the joystick ceases to move the focus box at all, recording or not.  I've looked for documentation on the feature, but obviously it's not in the user guide yet.

From what I've seen demo'ed, it could do with a softer start.  Looks a bit robotic.  Also, would be good to be able to set start and end points and speed as you suggest.

Thanks - I wasn't aware of this existing feature. I'll see if it works for me.

I only did a quick test last night with my my card Lexar 32GB 1000x and TL Oct 11 build, and I had no issues recording uncropped 3:2 video at 24 fps, with ~79.5 MB/s. I'll do more testing this weekend.

An idea - sorry if this was already discussed. Is it feasible to create a "virtual rail system" using crop mode on 50D? Are there any technical reasons which would only allow crop video to be sampled from the center of the frame? If not, perhaps such virtual panning is possible to implement? Say, one can specify how many pixels per frame the horizontal shift should be (could be negative), same for vertical shift. Then the crop area would be initially placed at one end of the full frame, and then would shift every frame by the specified X and Y pixels values, until it reaches the opposite end of the frame. I've seen this effect in some movies, looks very cool. I suspect it won't look real when using a wide angle lens (because of the perspective distortions at the edges of the frame), bur for 50mm and longer this should not very very obvious.

I am excited - finally I got my first speedy card (Lexar 32GB 1000x; ordered from US), so I plan to start playing with raw video this weekend (which happens to be a long weekend here in Canada).

Stock sites actually worth doing?

I have no idea. My photo microstocks site (dreamstime, were I already earned 600+$ for ~100 photos I have there) has just started accepting video. Contributors there are still trying to figure out what will sell and what will not.

Should be 7.6 minutes for uncropped video. And even that is an overkill for me (I'll probably only try to shoot microstocks video, which is limited to 30 seconds), but 32GB Lexars are apparently much faster than 16GB ones.

could be 8, I only have the 64s.

I think its possible to have B/W preview + no other indicators and GD on both on 50D and 7D. I didn't see any pink frames from cursory testing.. if you use anything else raw then pink frames or big slowdowns.

I am still yet to try the RAW video feature on my 50D. I was held back by the lack of a fast card, and by high cost of getting one (Komputerbay cards are not really attractive here, due to the ridiculously high shipping costs - 40$; given how likely the card would have to be sent back, this is a non-starter).

I ended up ordering Lexar 32GB 1000x - am I right that this is the card most likely to get the most out of 50D in terms of RAW video? That's what my research here showed anyway. I was thinking of 16GB, but it appears that the 32GB is substantially faster (145 MB/s vs. 95 MB/s rated write speed); I don't really care about the size (I doubt I'll do much video recording with my 50D).

These cards are currently on sale at B&H Photo video (until November 8 ): 67$ and 119$ for 16 and 32GB, respectively (comes to 87 and 147 CAD all in in Canada; the cheapest you can find in Canada is 218 CAD for 32GB, so it is big savings).

Is there a consensus on what build/trunk of ML currently provides the best uncropped RAW video capabilities? I don't think I am interested in the cropped mode, as I am a big bokeh fun, and only consider doing video with my 50D to make use of my fast primes (like Canon 135mm f2.0 lens).

Has anyone tried both Lexar 16GB 1000x and Transcend 16GB 1000x on Canon 50D? The Lexar is more expensive - I wonder if it provides any practical advantage when used on 50D?

What I find very frustrating when it comes to RAW video recording on 50D is the lack of information from ML developers about the most basic things - where to start, how does it work, which camera settings to use. I am good in photography, but novice in DSLR video, and I've been struggling for some time to get some answers here. It is probably here, spread in small bits across 109 pages of this massive thread, which makes it essentially unusable.

Anyway, I did some googling, and here is the number one place to visit if you are new to 50D RAW video:

http://www.eoshd.com/content/10770/canon-50d-magic-lantern-raw-review

This is a great review, aimed at novices like myself. For the first time I figured out what are differences, advantages and drawbacks for the full sensor and crop video recording. Now I know that the actual full sensor video resolution of 50D is 1584 x 1058 pixels, and it has some serious artifacts because of the way Canon does it. From reading other web resources I know now that how exactly DLR sensor is used for 1080p video recording is is not known - it is a "black box", fully hidden inside Canon firmware, or perhaps even done on hardware level. People can only speculate, based on actual video resolution measured from video, and amount of aliasing and moire artifacts. It looks like the best possible approach - binning sensor pixels into super-pixels for video - is not possible because of the limited bandwidth from the sensor. People speculate that some combination of binning and line skipping takes place, to reduce the bandwidth. This is really bad from a photographer's perspective - with huge DSLR sensors you'd expect to gain a lot in terms of light sensitivity (compared to consumer camcorders), but if you only do skipping, you don't gain anything - DSLR pixels are probably as small (or even smaller) than camcorder pixels.

People seem to be obsessed with crop video shooting, with the idea that the image quality should be much better, but as a photographer I can tell you that even the most expensive lenses (like my 135L) don't shine when viewed as 100% crops, especially wide open (and that's how I always use them - for DoF control). In other words, image quality is not great in crop video mode because of the lens limitations. One has to stop lenses down a lot to beat that. The second big negative coming from the crop mode is that you give up on DoF control (if you ask me, the main reason one should do DSLR video, and not consumer camcorder video) - because of the stopping down, and because of the fact that you are effectively working with a tiny size sensor (probably same size as camcorder sensor).

I wish this info, and more (like, details on shutter, aperture and ISO control when shooting raw video) where available somewhere as a single post, for novices. Then your great product would be appreciated even more.

H.264 is line skipping just the same as non-crop mode is with raw video then upscaled to 1080p. It's not resolving full 1080p. Crop mode records 1:1 pixels from a portion of the sensor. There is no inbetween capability. It's not adjustable.

re: Bayer pattern you'll need to address that to a1ex, g3gg0 etc. I'm pretty sure they know what they are doing and what these cameras are capable of. I'm not a dev btw

Thanks a bunch! I didn't realize 50D doesn't do true 1080p in h.264. And indeed, the 5D3 horizontal resolution is exactly 1920*3=5760 pixels (it shows the Canon's emphasis on video in the newest DSLRs).

So is this 3x3 pixels binning (from sensor -> to non-cropped video) built into the Canon's hardware or firmware, and ML is simply piggybacking it?

And is it indeed skipping, or binning? Binning would make much more sense - you reduce noise that way.

Still not clear what happens to mosaicing: Bayer cells are 2x2 pixels, and the video goes in 3x3 binning. Are there any demosaicing/mosaicing happening during RAWvideo recording, or only pure interpolation, separately in R,G,B? I think the latter would be more beneficial - fewer artifacts, and less cpu/memory intensive.

You need to be in crop mode to get 1920. Press the zoom button once then go back to the raw module menu. You'll see you can then select up to 2000 px

Thanks! And why is that? With non-RAW video, I can use 1920x1080 resolution. Is it because you hard-wired a certain binning factor - something like 1 video pixel for every 3 sensor pixels, in each dimension? Is it adjustable? I thought cameras like 5D3 can do 1920 in non-crop modes?

Given that Bayer cell has a size of 2x2 pixels, wouldn't the binning factor of two make more sense (should eliminate moire)?

Okay, I think I am starting to get it. I have to choose resolutions like 1500x1000 to use the whole sensor, without cropping - right?

But why can't I set 1920 pixels width, even when I use very low FPS (like 10)? When I try to do that, ML tells me "only 1580 is possible in this video mode"?

Still waiting for my fast card, but decided to play a bit with the slower (400x) Transcend I have. I installed the most recent TL build.

I am baffled with the zoom (crop) of the video. I don't see explicit settings for that. Say, I want to use the whole sensor (non-cropped video), 640x480 pixels. How should I do that? When I use grey-scale preview hack, my video becomes cropped by a factor of 2 or more. When I press the LiveView zoom button, it zooms in even further (does it affect the recording as well?), but pressing Zoom out does nothing - it stays zoomed in.

How can I do non-cropped video?

I think you have to de-bayer CR2 as well and ACR does it.

Where CR2 comes from? This is just for photos, not for video.

Can anyone please explain to me why there is even a need for demosaicing for the raw video? Especially when it is a non-crop (full sensor) recording? Demosaicing is only needed at the sensor pixel level, but with full sensor video, you are using multiple sensor pixels per one video pixel - so how come it is still mosaiced? Is it just to save on bandwidth and space (so instead of 14*3 bits per video pixel, you only store 14 bits per video pixel, sacrificing color resolution for the framerate and space)? And how is it accomplished - is the sensor data first demosaiced, and then a new (lower resolution) fake mosaic is created by ML (or perhaps Canon)? Or perhaps the raw sensor data is being interpolated/binned, separately for each channel (R,G,B), from sensor resolution to video resolution?

Thanks!

Dogmydog: Thanks, that is a good start.

My understanding is many here are using ACR as part of the workflow. That's what I always use for my raw photos, and I love it. One important reason I like is because it can automatically correct many lens defects across multiple photos - distortion, vignetting, and chromatic aberration. CA is probably not very important for video, but distortion and vignetting are.

In your workflow, at the ACR stage, does it get all the relevant meta data (EXIF)? Is there any meta data in the RAW video? At the very least, one would want to have the lens ID, aperture, and focal length (for zooms) at the start of the video. Better still to have it for each frame. Are there any plans for such feature, if it doesn't exist yet?

I remember some time ago someone posted a link to a guide/tutorial for RAW video on 50D, but I can't find it. Could anyone please repost it? It should probably go to the first post on this thread - it became so long, that for someone in my situation (about to start using 50D for RAW video) it will appear overwhelming.

Thanks!

@pulsar

whatkind of powersolution u have there?

I haven't received the card yet. I only tried compressed 1080p video on my old Transcend 400x, works fine.

I can only blame this thread for quite a bit of unplanned expenses !  At this point, I haven't shot any RAW video yet, but I already spent money on




• - Zoom H1 recorder (I'm still hoping for a ML solution for video/audio syncing, like making 50D beep at the start of the video...)
• - Zoom H1 accessories (shock mount from dslrnoob, wind shield)
• - Power adapter for my 50D
• - 200W power converter from 12V to 120V (to be used with the power adapter and the car starter battery I already have)
• - And the latest - I just ordered locally Transcend 32GB 1000x. For Komputerbay 32GB 1000x, shipping to Canada and duties make it as expensive as the Transcend, and from what I've seen the Transcend should be much more reliable.
  

I am itching now to put all this to good (RAW video) use. I have a few nice lenses to play with (Sigma 17-50 f2.8, Canon 135L, Canon 70-200 f4L, Sigma 10-20mm, Samyang 8mm fisheye, 50mm and 28mm MF lenses).

I've also ordered the Transcend (locally), as shipping of Komputerbay cards to Canada kills any price advantage (one has to pay 40$ for shipping plus whatever duties and taxes and processing fees they'll come up with; this doubles the price for the 32GB version). From what I've seen, Transcend should be much more reliable than Komputerbay, and perhaps slightly faster.

[without recoding]

Not exactly on topic, but I'm trying to learn 50D movie recording using the old non-RAW (compressed 1080p) video hack, which produces MOV files. If anyone here did shoot video this style - what is your workflow? I have an external recorder (Zoom H1), so the workflow should include a way to demux MOV into h264, sync it with external WAV audio, and mux it into something like MKV or M2TS. Is there a tutorial somewhere which would explain every step? Also, may be there is a tutorial on shooting non-RAW video with 50D - what settings to use (like, picture profiles etc.) and how?

Finally - it is probably impossible, but perhaps there is a way to join compressed video produced by 50D (which is 1080/30p h264) with the output of my Canon HD camcorder, which produces 1080/60i h264 - without recoding? I think there are containers out there which allow video stream to switch frame rates, interlacing and non-interlacing, in the middle of the stream. May be MKV can do that? Which video tool would let me join non matching frame rate files? I tried a few (AVdemux, tsMuxerGUI etc.) but it doesn't work.

The initial responses were sort of positive, but it's been a while without any feedback.

I know my 50D can beep - with AF focus confirm. It is not loud but it might be good enough for audio syncing with on top of the camera audio recorder (like 100$ Zoom H1 - BTW I just got a cheap shock absorbing mount for it from video noob store, works well), in most situations (unless it is really noisy).

All this RAW video stuff for 50D is cool, but will not be very useful without a good audio syncing solution.