Panasonic's AVCHD-Lite breaks MeGUI, DGtools

[trevmar]

Panasonic's new ZS3 camera has implemented a new version of AVCHD which Panny calls "AVCHD-Lite." It uses a 60p video stream, which messes up many of the AVCHD codecs out there. But the new ffdshow revision 3054 will at least play the raw video in Media Player Classic (this new ffdshow is in K-Lite update 508_20090823).

However, I cannot process it to another format using MeGUI, and any attempt to create an indexed .DGA file also fails. What happens is that MeGui produces a video with 60p frame rate, but the video content is repeated twice during the clip. Same when a .DGA is created - the video repeats.

I have attached an 8 second clip of raw AVCHD-Lite video from the camera at URL:
http://trevormarshall.com/temp/00001.mts

The AVIsynth control file used for MeGUI contained just the line:
DirectShowSource(".\00001.MTS")

and it did display correctly on the video preview window which MeGUI pops up. But there is clearly some sort of frame rate issue with the 60p AVCHD-Lite frame rate.

I am assured that the camera plays these files correctly through its HDMI output, but I haven't attempted to do a capture of that video to see if it is AVCHD-Lite or AVCHD which is being fed to the TV set.

Has anybody any ideas that would help me get a video-conversion tool chain running for this camera output? At this point I am not even sure whether the camera output is 60p or 30p, as the motion blur doesn't look right for 30p. But none of my usual tools handle the video. I would normally feed the video into Virtualdub with AVIsynth, but AVIsynth also fails to handle the AVCHD-Lite properly (which may be an ffdshow, issue, I can't tell).
.

[poisondeathray]

AVCHD lite has frame duplications (every 2nd frame is duped). The actual unique content is 30p. Most decoders and splitters have not been updated yet for compatibility

FFMS2 will return the proper 30p without the dupes, you can use TSMuxer to demultiplex the audio and feed it in separately

FFMpegSource2("00001.mts")

or if you want audio in the same script

FFMpegSource2("00001.mts", atrack=-1)

Cheers

[trevmar]

Excellent, yes ffms2 works fine to allow AVIsynth and MeGUI to process those files. Thanks for the help

I have a nagging desire to be able to see the dupes, however, a desire to step through the raw 60fps frames one by one. I did manage to use Gunnar Thalin's 'Dinterlacer-smooth' with CoreAVC to reveal that dupes were apparently present, but that was a kluge.

I couldn't find anything in the ffms2 manual about a switch to also process the duped frames into the output. Is there an option such as this, or something else I could use?

[aedipuss]

it seems to really be 60p. vegas pro9 correctly identifies it and uses it.



2009-08-31_123756.jpg



zs3.mp4

[poisondeathray]

You can use DSS2() with haali splitter and ffdshow to see the dupes

@aedipuss - It's 60p with dupes. So really 30p for unique content. Using a stream analyzer on the raw avc stream confirms this. Vegas has the same problem as most software with these streams. If you disable resample (right click the clip on the timeline) , vegas can use this properly. This has been discussed before on another thread

[aedipuss]

vegas uses it ok without disabling resample. i can export as 30p -no dupes or 60p with the dupes.

[poisondeathray]

If you examine your vegas encoded file you will see blended frames

Vegas mistakes the file as 60p (like most software)

The musicans are horizontally still, so it's not perfectly evident, but look at the cars in the background

vegas
http://tinypic.com/r/10fuecn/3

ffms2
http://tinypic.com/r/oa915u/3

Other AVCHD lite cameras exhibit this same behaviour with vegas

Cheers

[trevmar]

.
DSS2 works great - thanks

I am happy now My toolstream is under control again, even if some of my codecs (eg CoreAVC) are not updated yet.

Thanks again, poisondeathray
.

[creamyhorror]

I've read about this duping, but I don't get why the camera does it. Doesn't AVCHD-Lite allow 720p30? Is there some technical reason to do it?

[trevmar]

.

I've read about this duping, but I don't get why the camera does it

My guess is that it makes it easier for them to simplify the HDMI output socket hardware, but that is just a wild guess
.

[poisondeathray]

I've read about this duping, but I don't get why the camera does it. Doesn't AVCHD-Lite allow 720p30? Is there some technical reason to do it?

It's a marketing ploy. Yes, you are getting 60p (which sounds better than 30p) but the fine print really means 30p unique frames.

It's just like 30i marketed as 60i

I think there is a flag in the stream signalling to play at 60p with dupes, not sure exactly on the mechanics. But most software will misidentify these. I'm sure as this format gets more prevalent, proper support will be added to programs. If you recall there were similar issues with AVCHD when it first came out

[trevmar]

It's a marketing ploy. Yes, you are getting 60p (which sounds better than 30p) but the fine print really means 30p unique frames

Proper motion blur is always a problem with still cameras, as they cannot emulate the CCD/CMOS readout schemes used by (for example) the HF100 to give smooth film-like motion blur at 24fps (and 12 fps for that matter). The cameras give discrete fast-shutter snapshots.

Surprisingly, the panny ZS3 has a similar (within 2 f-stops) low-light sensitivity and noise to my HF100 at 24fps. Which is quite exceptional. Only the v1073 has done better for me: http://www.vimeo.com/1028849

I can add motion blur back during post-production. I usually use the "Dynamic Noise Reduction" filter in virtualdub (and as an add-on filter for Vegas). But I can't bear to lose control of my raw format, or allow de-interlacing artifacts to reduce resolution prior to my own messing with the images So thanks for the tips
.

[aedipuss]

If you examine your vegas encoded file you will see blended frames

Vegas mistakes the file as 60p (like most software)

The musicans are horizontally still, so it's not perfectly evident, but look at the cars in the background

vegas
http://tinypic.com/r/10fuecn/3

ffms2
http://tinypic.com/r/oa915u/3

Other AVCHD lite cameras exhibit this same behaviour with vegas

Cheers

hehe - thanks for pointing it out. i would have missed it - those ghosted extra wheels on the cars are about all that gives it away.

[creamyhorror]

It's a marketing ploy. Yes, you are getting 60p (which sounds better than 30p) but the fine print really means 30p unique frames.

Duplicating frames counts as a "marketing ploy"? I'd say that's outright fraudulent advertising. Worse, it's causing software to misidentify the video, as you said.

And isn't it killing the bitrate efficiency of the footage? They're damaging their camera's recording times!

[poisondeathray]

And isn't it killing the bitrate efficiency of the footage? They're damaging their camera's recording times!

No efficiency lost because the stream itself is 30p when you analyze it (i.e. there is no physical duplication). I think it's a flag or some signalling in the stream that is causing most software to repeat frames upon playback.

But the streams do have lower efficiency, even worse than AVCHD - (they have no b-frames)

[creamyhorror]

No efficiency lost because the stream itself is 30p when you analyze it (i.e. there is no physical duplication). I think it's a flag or some signalling in the stream that is causing most software to repeat frames upon playback.

Oh, okay. Even more ridiculous to call it 60p when all the camera does is set a flag.

Tangentially, when dealing with R.A.J.'s footage from his Canon videocamera in that other thread, I ended up with an incorrect (double) playback length in my x264 encode, after NNEDI2(). I was using FFVideoSource to import the footage. Was that a similar problem to the one in this thread, or should I have been using another method?

(I was going to try AssumeFPS(25) in the AVS script but didn't bother in the end. Thought it might fix the signalled framerate, which seemed to be 50fps as reported by x264 during the encoding.)

But the streams do have lower efficiency, even worse than AVCHD - (they have no b-frames)

Okay, now that's really terrible. Even XviD has B-frames. What kind of AVC encoder is that??

To think I was considering getting a ZS3/TZ7...I wonder if most consumer HD point-and-shoot cams are like that? Will have to check some raw footage tomorrow.

[trevmar]

(they have no b-frames)
Okay, now that's really terrible. Even XviD has B-frames. What kind of AVC encoder is that?? To think I was considering getting a ZS3/TZ7...I wonder if most consumer HD point-and-shoot cams are like that? Will have to check some raw footage tomorrow.

The sad news is that video from the other P&S cameras is equally bad or worse, with the exception of the (large) panny GH1 and Canon DSLRs. The ZS3 was the only pocketable I could find with any vestige of manual controllability in video, and with the ability to record for longer than 30 minutes at a time...

On hearing about the B-frames I did change my video speed selection from H to SH to compensate for the lack of B-frames. Wah - I only get 2hr 05 minutes on a 16 gig SDHC at the higher bitrate...
.

[poisondeathray]

Oh, okay. Even more ridiculous to call it 60p when all the camera does is set a flag.

I'm just guessing that's the mechanism (I don't know for sure), because the raw stream is 30p

Tangentially, when dealing with R.A.J.'s footage from his Canon videocamera in that other thread, I ended up with an incorrect (double) playback length in my x264 encode, after NNEDI2(). I was using FFVideoSource to import the footage. Was that a similar problem to the one in this thread, or should I have been using another method?

(I was going to try AssumeFPS(25) in the AVS script but didn't bother in the end. Thought it might fix the signalled framerate, which seemed to be 50fps as reported by x264 during the encoding.)

ffms2 is buggy like that, you sometimes have to use assumefps, or other source filters. For example on that horse clip DSS2() works fine

But the streams do have lower efficiency, even worse than AVCHD - (they have no b-frames)

Okay, now that's really terrible. Even XviD has B-frames. What kind of AVC encoder is that??

Maybe that's part of the reason they call it "-Lite" ? (along with the bitrate & resolution differences)

I wonder if most consumer HD point-and-shoot cams are like that? Will have to check some raw footage tomorrow.

Like trevmar said above, most point/shoot cameras are actually worse. They usually use baseline level AVC wrapped in .mov container, or -gasp- mjpg in avi

[creamyhorror]

The sad news is that video from the other P&S cameras is equally bad or worse, with the exception of the (large) panny GH1 and Canon DSLRs. The ZS3 was the only pocketable I could find with any vestige of manual controllability in video, and with the ability to record for longer than 30 minutes at a time...

That's true, the cameras I'm looking at (Canon SD780 and Samsung HZ10W) don't seem to have bitrate/quality options. At least the Canon records at like 20Mbps though. The Samsung records at 9.8 I need to go look at the streams from those cameras more carefully, though I suspect their lenses don't produce all that spectacular video anyway.

ffms2 is buggy like that, you sometimes have to use assumefps, or other source filters. For example on that horse clip DSS2() works fine

Didn't realise that, good to know.

Maybe that's part of the reason they call it "-Lite" ? (along with the bitrate & resolution differences)

I assumed Lite just meant 720p-only devices (which is what Wiki said). Don't know if it has any restriction on bitrate and encoding settings beyond AVCHD's.

Like trevmar said above, most point/shoot cameras are actually worse. They usually use baseline level AVC wrapped in .mov container, or -gasp- mjpg in avi

No CABAC, no b-frames...you might as well just encode to DivX. (My old Casio Exilim did that, and it was very nice and space-efficient for its time...2006.) MJPEG still plagues modern PnSs like a bad disease.

What useful tools are there to glean information about AVC streams? I know about detecting frametype and coded size from ffdshow's OSD, and of course use MediaInfo all the time, but are there other methods to get info short of buying Streameye?

[poisondeathray]

What useful tools are there to glean information about AVC streams? I know about detecting frametype and coded size from ffdshow's OSD, and of course use MediaInfo all the time, but are there other methods to get info short of buying Streameye?

H264Visa , it's not free either and a bit more hardcore than streameye. Lord Mulder says the "trial" is never ending (supposedly you can uninstall/reinstall unlimited times) , but if you do use it that much then consider buying it.

[trevmar]

.
Oh - and the inbuilt audio on the ZS3 ain't half bad (as you could hear from my clip).
.

[creamyhorror]

H264Visa , it's not free either and a bit more hardcore than streameye. Lord Mulder says the "trial" is never ending (supposedly you can uninstall/reinstall unlimited times) , but if you do use it that much then consider buying it.

I forgot I downloaded and tried using that. But the price is...pretty extreme, IIRC. Will try it anyway.

trevmar: Haven't seen your clip yet, but stereo recording is no doubt a good thing.

If only cameras were more hackable. I know the Canon firmware hacking project (CHDK) has managed to give users multiple quality (bitrate) options on certain cameras. Other brands don't seem to have anywhere near the same level of firmware modification, though. Would be real nice to be able to choose your bitrate on cams that don't support it.

[MJ Peg]

It looks like the AVCHD spec

http://en.wikipedia.org/wiki/AVCHD
only allows for 60p/50p/24p frame rates at 1280x720, so if they could only manage 30p or 25p in the camera then they had no option but to set the flag for each frame to be shown twice. Otherwise it couldn't be called AVCHD. At least it's compatible and works well with TVs and DVD, etc.

The file sizes and bit rates indicate that only 25.0/29.97 frames per second are actually stored, so anyone worried about frame 'duplication' in the bitstream needn't worry.

I have converted the '50p' clips to AVI with 25p and they play identically on the PC.

[creamyhorror]

[quote:5766d3cc60="MJ Peg"]It looks like the AVCHD spec

http://en.wikipedia.org/wiki/AVCHD
only allows for 60p/50p/24p frame rates at 1280x720, so if they could only manage 30p or 25p in the camera then they had no option but to set the flag for each frame to be shown twice.[/quote:5766d3cc60]
Good point. I guess it's to preserve compatibility (in some way) with Blu-ray specs, which also don't allow 30p. But I don't understand why they left 30p out of both in the first place, given that it's a natural recording format for the current and upcoming generation of consumer cams. I can't help but think that they were trying to suppress personal Blu-ray authoring (think I read that somewhere).

[ChrissyBoy]

The ZS3 was the only pocketable I could find with any vestige of manual controllability in video, and with the ability to record for longer than 30 minutes at a time...

Is this true? I have a TZ7 which i believe is the UK version and that only does 15mins. Have i missed something?

[trevmar]

[quote="ChrissyBoy"]

I have a TZ7 which i believe is the UK version and that only does 15mins

My ZS3 records continuously for as long as the battery lasts, or about 90 minutes of non-stop seamless high quality AVCHD-lite video.

[ChrissyBoy]

What an arse. It seems that non EU versions are indeed like that. Whereas we in the EU have the 15 min limit!!!!:

[MJ Peg]

and yet it's a 29m 59s limit on the FZ38 for that same EU tax reason.... looks like Panasonic could have made it longer than 15m

[ChrissyBoy]

Well not wanting to hijack this thread i will just say that it is perplexing for features and tax to be releated..... If they priced it £10 more for unlimited recording - surley everyone would be happy?

[MJ Peg]

Google found me pages that say that "European Tax law applies 4.9% import duty on camcorders made outside of the EU" ... the TZ7 is about £250 so it's a whopping £12-ish

Yet, it's a competitive market and that extra £12 quid could make all the difference and hurt sales!