Frozen frames with 600D/T3i: Canon, ML or SDHC to blame ?

[benoitvm]

Made some holiday video with my new EOS 600D (T3i) and ML. Using Class6 SDHC cards, as recommended by Canon. Default bitrate.
On some occasions, I had the "five-level indicator" appearing shortly before the recording automatically stopped. I understand this might be due to the fact I was shooting "difficult" scenes (pan across "busy" landscape with lots of details, leaves, etc..), and the instant bitrate (default is VBR recording on this camera) was probably peaking too high....

But the real problem is that in some other clips, without apparent reason, and without any notice during recording, I have occasional "frozen frames", i.e. 4 identical frames (meaning 3 lost frames). There is no interruption in the sound, and the sequence of the individual frames suggests that there was no interruption in the video recording either, just short bursts of "frozen frames" (each time 4 identical frames). Any clue what this could be ? A bug in the AVCHD encoding (Canon...), a problem with the unit, or a side effect of ML (eating up some valuable CPU cycles...) ? Needless to say, it is quite annoying: it is impossible to detect before carefully watching all clips after shooting, and the cure, if reshooting isn't possible, is to try some frame interpolation technique in the editing phase....

Note: the Class 6 SDHC card used works fine with Canon HF-G10 AVCHD Camcorder at highest bitrate (24 Mbps), and is even recognized as being class 10 by the camcorder.

[a1ex]

What version? Can you reproduce?

There was a similar issue reported by a 5D2 owner. I've looked at the video frame by frame and could not notice any dropped frames. Can you post a sample video?

[benoitvm]

after some new testing back home, it appears that the "freeze" occurs and is visible also in standby mode, so the codec is just encoding what it receives from the camera image processor....and it happens also when a ML-free SD card is used.....
The latter test was done with the EFS 18-200 mm (set to 18 mm) with Stabilizer set to OFF...
So I would think it is a camera (general or unit) issue - the best thing is probably to have the camera checked (and repaired ?) 

But first I will try to test with another lens.

[benoitvm]

same problem with the EFS 18-55 lens....

[a1ex]

Clear camera settings and show us a video of the problem.

[benoitvm]

https://dl.dropbox.com/u/2690810/EOS600D/MVI_1309.MOV
(720/50p) - shot with ML disabled
two occurrences of 5 identical frames, near 3.5" and 6"

I also have some clips from my holidays, with less obviously "difficult" scenes and the same issue, in 1080/25p, but I prefer to send the link privately, if you need them (let me know which email address I should send to)

I also noticed that the camera is producing a kind of little mechanical clicking noise during live view or recording exactly when the problem happens. Wondering what that could be, since I set IS to off and focus to MF ?

[a1ex]

Downloading. Are you sure you are using manual exposure mode?

[benoitvm]

no, auto exposure (for a number of reasons out of this scope) - would that matter ?

[1%]

and it happens also when a ML-free SD card is used

I also noticed that the camera is producing a kind of little mechanical clicking noise during live view or recording exactly when the problem happens.

Sounds like a HW problem to me. I cannot reproduce (thank god!). Sensor or processor is freezing/depowering and encoder re-writing the same data? Light clicking would be shutter closing. Anything overheating?

a side effect of ML (eating up some valuable CPU cycles...)

I have "unstable" memory allocation in my  TL builds and nothing like this ever happened. I've had the camera die with the sensor and LV still on but everything else not; so the opposite situation (not since june).

Is there any way you can consistently reproduce the problem?

[benoitvm]

Is there any way you can consistently reproduce the problem?

consistently no, but easily yes: turn on the camera, and start doing fast pans (recording or standby) ....

Before jumping to the store for repair, I will try removing ML and upgrading Canon firmware 1.0.2  (current ML only supports 1.0.1 on this cam, right ?). Will post results here ASAP.

[a1ex]

Do try in manual mode.

[benoitvm]

Good suggestion !

Made some progress: in manual mode, no issue, even when I (manually) change aperture during recording (the little clicking noise is heard, but no frames are dropped). Clearly, in auto mode, it is the automatic aperture change (-> clicking noise) that causes frames to be dropped (resulting in frozen video). So it turns out that it is not the quick change of the content of the filmed scene (like fast panning), or its complexity, but the changes in lighting condition that causes the problem.
Can anyone repro this problem in auto mode ?
If this is a general issue with this camera type, I understand serious enthusiasts never use auto exposure, but in the case of casual filming of unpredictable on-the-fly scenes perhaps with difficult lighting conditions (holidays), auto exposure can come in handy when you don't have the time to properly setup the camera...Why would auto exposure be plagued by such a limitation ?

[benoitvm]

Just FYI, I just removed ML and loaded Canon firmware 1.0.2 : no change !

[scrax]

Just FYI, I just removed ML and loaded Canon firmware 1.0.2 : no change !

also after resetting camera settings?

[benoitvm]

Yes indeed

Just waiting for someone somewhere to repro before returning the unit to the store 

[1%]

Try the same settings from auto mode that froze it in manual mode?

[benoitvm]

To be clear:
The problem happens (apparently) only in Auto mode
I could not repro in manual exposure video

The problem is independent of the firmware (Canon 1.0.1, 1.0.2 or ML)

[benoitvm]

Sorry to bother you this forum...since it looks unrelated to ML
So nobody could repro this problem ?
Did anyone actually try ? (before I return the camera for repair)

[benoitvm]

I went to the store today and it looks like we could repro the problem with another 600D...
So I will contact Canon support here in Europe, but if anyone has additional info (or can/cannot repro in various circumstances), it could still be helpful. Tx

[a1ex]

Use a manual lens

[benoitvm]

Use a manual lens

Agree, manual exposure is what most if not all serious DSLR videographers do, but there are situations where you don't have the time/opportunity to properly manually expose your shot. It was the case for some events during my holidays and this "feature" ruined several unique shots...

[ilguercio]

Send a mail to Canon, try to see if they have an answer.
It might require a fix

[benoitvm]

Sent an email to Canon.

In the meantime, I finally found one other user facing this exact same issue: http://www.eos-numerique.com/forums/f102/choix-dobtique-s-je-narrive-pas-me-d-cider-183247/#post1970875235 (sorry, it's in french...)...And it's with a 550D !!!!

Note for those who want to try to repro: the trick is to have simultaneously strong lighting change AND fast subject movement; strong lighting change is required because apparently, in AUTO exposure, the EOS seem to favour ISO and shutter changes prior to changing the iris aperture (and it is this iris aperture change that causes (or comes together with) the symptom); fast subject (or camera) movement is required in order to catch this <= 1/4s freeze....

[1%]

Filming something and accidentally took a picture. 3 frozen frames ended up in the video.

[benoitvm]

Following my email, the regional Canon support center (France for my country) has called me to explain that what I am seeing "is normal because of the H.264 compression technique used, which uses keyframes when the picture changes significantly and hence a few identical frames are following the keyframe, etc..." (or something very similar)

As I insisted that I do understand how H.264 or GOP-based encoding generally works, and that this kind of answer is absurd (and plain wrong), the engineer added that "all cameras using H.264 exhibit the same behavior"...I asked: "all Canon cameras, but also Nikon etc... ?", he replied "yes, it is due to H.264", challenging me to go to a store a see by myself.

Needless to say, I won't waste time for such a stupid test, being sure that only some Canon models are affected, most probably those based on the 550D family, and that the problem has nothing to do with H.264 encoding (the engineer maintained his explanation, despite me clearly explaining that the problem occurs regardless of the camera recording or being just in standby). I asked to have this explanation by email, if possible straight from Canon engineering (Japan). Nothing has arrived in my mailbox yet....

The guy on the phone concluded that "since this is inherent to H.264, no firmware fix for such a problem could ever be developed". 


Just to be sure: do you guys agree that if the problem is visible on the LCD (or attached PC) also when the camera is in standby, it means that the cause is upstream of the H.264 encoder ?