canon 5d mkii and compression artifacts in the orginal footage
 

Hi,

I just finished shooting for 5 days with the 5d mk2. We were using the magic lantern 'firmware' software on the camera. We shot on to a number of different CF cards: sandisk extrteme iii and extreme iv (8gb) as well as kingston 266x (8gb). On a few of the days (not everyday!) I got a lot of compression artifacts in the footage- it looks like some pixels get stuck for half a second around motion, such as you would see in a really badly compressed DVD or web video. This is on the footage straight out of the camera!

After the first time I noticed this, I re-formatted the cards using disk utility on my mac (even zero'd them out!), but it didn't help.

I can't seem to find anything on the internet about this issue, so I'm wondering if it is only this particular camera. We bought it just a few days before the shoot. Maybe it is defective and should be sent back?

As far as I can guess it is either:

-defective camera
-some problem with the formatting of the CF cards
-an issue with Magic Lantern software.

Has anyone come across this issue?
Thanks.

Liam

Some people feel that in complex scene with deep dof the compression engine can't keep up. I questioned at the beginning of magic lantern if the CPU could give up the extra processing to the added ML features.

Were the artifacts in visually busy scenes? I use external audio and a Marshal monitor that has focus and exposures aids, so in general I don't use ML.

Magic Lantern may very well turn on functionality that Canon felt needed to be left off. But for people without compression problems there's no reason not to use ML.

Unfortunately it's unlikely Canon will be able to provide quick help. Shallower dof is safer as far as artifacts.

Are your extreme III's the newer version? The cards need to be UDMA.

As far as formatting, I would recommend doing that in camera. That way you know the card is set up according to Canon specs. I don't know if that can be a problem, but why not eliminate it as an issue.

I just shot a film this weekend with the 5D, using Magic Lantern, depending on circumstances. I have not seen the issues you are indicating.

One known bug of Magic Lantern cost me a scene. Sometimes in Magic Lantern, when I start the first shot after shifting into ML, the recording will stop after a 3 pr 4 second run. I have gotten myself into habit of starting and being sure it is running. In this case, I did not, and the result was I only had 4 seconds of footage for the take.

I've never seen artifacts like you describe, while it could be a bug caused by Magic Lantern, my guess is a card problem. As Don Miller pointed out, if you are not using UDMA cards the camera will degrade the quality of the recording. This info comes from Tim Smith's talk at DCS.

I get that with my Kingston 133x 16 GB and my Sandisk Extreme III 8 GB cards, with or without ML running. I think it is a problem with the cards -- there were lots of people reporting problems with the Sandisk cards manufactured around November of last year. This bug almost always is the first take after a "reboot".

That is completely normal. The compression artifacts get worse with more fine detail and motion. The way that most of us deal with it is to use thin DOF so that there is very little fine detail in the image.

I had a similar experience shooting a busy scene off my deck this weekend. I had only had the camera for 2 days, and that clip ranks as one of the ugliest pieces of video I have shot in 5 years.

I had a very sharp 17-40 F4L on the camera. I went out on my deck with the lens set at about 20 mm, aperture at maybe f8.0. The scene from my deck is a river valley with about a two mile view of pine forests, cliffs, willows, and grasses. I was trying for a handheld wide-angle deep field of view clip. There were what appeared to be CMOS wobble, aliasing, and codec stress artifacts all over the place, so much so that half of the objects appeared to be rapidly cycling in and out of focus. Throw in the distortion this lens has at wide angle, and it looked just like a scene from one of my worst migraines! I have a UDMA card and I am not using ML.

Other scenes I've shot look okay to good, I'm still learning. I would tend to agree that you have to be careful with high detail, deep depth of field, and motion all in the same scene. I learned from folks on this forum and others to dial down contrast and saturation in the neutral picture style (now contrast and sharpness are at their lowest setting), and this seems to have improved resolution and decreased artifacts in the files.

Pat

Is the shot repeatable (with the same artifacts)?

Extremely unlikely to be the cards. Those cards should have a lot of headroom for the data rates anyway, and I'd expect them to either work fine or fail altogether. (As happens with duff SDHC cards in an EX.) What you suggest implies that if they are too slow, they force the camera into a more compressed/lower bitrate mode. I don't think that's very likely.

The compression being used is pretty heavy - hence why you can get what you can on the cards. I'm afraid it may be an unavoidable issue with the camera, and all you can do is what Daniel suggests.

My experience
 

At first, I have been quite displeased with the artifacts described here. Still, I have started to use 5D2 footage professionally (for broadcast) and have no problem with it. With some restrictions: as little as possible details, as little as possible movements (I am generally not moving the camera anyway, that is my preference). Generally, shallow DOF provides images with no or with acceptable artifacts. Narrow rather the wide angle lens is best used. However, I use 17 - 40 mm quite a lot, but always in situations with not much details. Walls and other plane surfaces will enable good quality images even with wide angle lenses. Wide lens + lots of details = unacceptable image quality. Add movement to this equation and you will wish to slam your 5D2 against the wall. Or pavement.

Also, since I am in the PAL area, the 30 fps issue I tackle in the way that I do not alter the footage to 25 fps if I do not have to (provides nice and smooth atmosphere I like, when played on 25 fps timeline), and when I have to alter the frame rate to 25 - hey: my shoots are still anyway, so very rarely one can detect any change. When that happens - and it did only couple of times - I simply do not use that particular shot. When shooting, I always take much more then needed, so there is always alternative in my rushes.

So, once you get to know strong and week points of your tool, it actually becomes fun. I do not take my EX3 to some shootings at all now. And love it. And using 5D2 in some situations in documentaries, enables you to take footage you would not have been allowed to take with video camera. Finally, I transfer Canon footage to XDCAM EX codec and edit just fine with the rest of my shoots taken with EX3. Made several screenings for my colleagues, some recognized shots been taken with Canon, some did not. But, non of them had any complaints. On the contrary!

Final remark: among the audience I made test screening for, the guys who are producing films on regular basis never detected any difference! And guys that detected the difference are AS A RULE the guys that talks much more then film (no offense meant). My explanation is: the first are driven to the content, the second to appearance. Go figure.

This is only my experience. I thought it might be interesting to the ones that just started to use 5D2.

I'm with you, I'm just not buying the whole UDMA requirement thing. The implication in some of those statement was that bit rate or something is changed on the fly when it detects a slower card. I call BS on that. In my experience/opinion, if you aren't seeing the buffer fill up on the screen, then you're getting everything the camera had to offer. And since this is digital, you're either getting the file being written or not.

It is hard to get a handle on the complaints here, without seeing the actual "artifacts". I also wonder if the OP might be having issues with a playback device. I have seen some bad artifacting and blocking on Media Player Classic which seems otherwise to be able to play the files full screen on my system without stutter, with an admittedly poorer image, than I can get with Quicktime. I find that some so called artifacts are minimized or disappear on conversion to my intermediate codec with NeoScene.

As far as Pat's comments, this camera should never be confused with a hand holdable video camera. If you are going to move with it, you need a very steady platform, or you will get a mess. Because of the design, what you think of minimum movement will be translated to the various axis of the camera in ways we are not used to in the normal "long" and narrow video camera orientation. In my testing, a shoulder brace gives a lot of that stability back. Otherwise by adding that motion, and the fine detail attempted in a wide angle deep depth of field shot in the first place, I have no doubt you will tax the codec capabilities to its limits.

This is one frame out of a vertical tilt without a tripod:
http://www.digipiction.de/grade/terrible.jpg

Truly unacceptable and no detail whatsoever, right?

Seriously, there are NO visible compression artifacts on unaltered footage (the above scene is heavily graded by the way). I'm not sure why some people don't understand how much headroom 40Mb/s H.264 gives.
H.264 at these bitrates is so good that ANYONE can do a single-pass encode and not produce artifacts. Just try it out on some uncompressed footage.

Hi again,
Thanks for all your responses. Sorry for the slow response- I've not been around a computer. So I guess there is no real solution? It's just a limitation of the camera? We're doing a lot of hand-held stuff (using a cavision rails rig) and I use the 16-35 2.8 quite a lot... bad news for me! Here is a link to a frame grab-

http://liam.idyllpictures.com/5d_glitch.jpg

it is a sever example of the problem I'm talking about, though not unfortunately, not a rare one (sorry not to imbed it- I don't know how to do that). The jpg was taken from the original file off of the card- it has not been processed at all (other than turning it into a jpg!)

The camera was on a jib arm, but it was not moving much at all at the moment of the glitch- just the people in the frame were moving.

It can't be a play-pack issue, as it is always the same frames and even when going frame-by-frame in FCP it is there. The problem is the same after encoding to prores too.

As to the cards- they were bought about a month ago, and there were artifacts even on the extreme IV (45mb/s), I guess I'll have to check if they are UDMA...

as to:
"This bug almost always is the first take after a "reboot". (from Tramm)"

Would it help to roll for a few seconds on some nothingness and stop before actually shooting?

Another question- is there any after effects/motion/FCP plugins that might be able to repair such footage?

Thanks again

Liam

...I guess I'll have to keep waiting for a Scarlet!

That is not a compression artifact - that's some sort of error in the file. I've never seen anything like that in any of my 5D footage, and I'm using both Ultra II & III cards (not using Magic Lantern though). If you're seeing that regularly I would suspect something is wrong with the camera.

I haven't seen anything like this ever (and the last piece of footage I shot was called MVI_5903-001).
Either your camera or the memory card are probably defective.

P.S.: I've never used Magic Lantern

hmm... looks like a trip to bangkok to see canon is in order...

Sorry for the delayed response. I haven't seen any artifacts remotely like that either.

I haven't tried to duplicate the problems I was seeing on a landscape shot with the 17-40 F4/L, but I have managed to improve the quality of the deep focus landscape shots with this lens. Now I am using the neutral picture style with sharpening and contrast all the way down and saturation down a notch. I am using Cineforms Neo HD and and First Light to convert and color grade the clips with final processing and sharpening in Vegas. I have eliminated most of the CMOS wobble with better camera handling techniques. The resulting wide angle shots look quite good.

Ironically, landscape shots with the 70-200 F4/L, 135 F2/L, and 300 F4/L IS look relatively soft and full of artifacts no matter what I do. Focus is spot on and checked with magnification in live view. Apertures are set to nearly ideal for sharpness in each case. I use low ISOs, have checked exposure, and set the shutter to 60 per second. It doesn't matter whether the camera is still or I slowly pan (on a tripod), the details are pretty soft (nothing in the frame but distant landscape features). In the same situations the still images are incredibly detailed and the Canon XH-A1 clips look great. Using the same lenses on the 5D II for closer subjects I get better detail, although not quite as much as with the XH-A1.

Consistent with many others' tests and observations, I've found the XH-A1 to have better effective resolution. Fortunately I never planned to use the 5 D II as a replacement for this camcorder. I gave up still photography 4 years ago when I bought the Sony HC-1. The 5D II gives me an opportunity to get back into it. In the next few weeks I hope to shoot some informal music videos in dark clubs with the 5D II, and learn how to take advantage of its strengths as a video camera.

Pat