GH1 firmware research volunteers required

What sample where you looking at ? Perhaps you are desribing something else, not keyframe pulsing ?
I'll post a clip. No doubt in my mind that it's keyframe pulsing.

because in theory a P-frame cannot be higher quality than the I-frame it's using as reference.
Totally disagree. Of course a P-frame can be much higher quality than the I-frame. The P-frame is not targeted to match the I-frame, it's targeted to match the raw uncompressed frame that's fed to the coder. A P-frame looks at the changes between the previous frame, and the target frame. If the target frame is incredibly complex, and the previous frame is mushy, the P-frame will use its bandwidth to encode the changes between the mushy frame and the detailed reference frame.

Each frame in the GoP does not care what the prior frame looks like, it instead cares what the target frame (the uncompressed image) looks like, and makes its best effort to look like that target frame. So yes, cyclic codec pulsing is very possible if the I-frame does a poor job on the first frame in the GoP.

In the clip I will post, you can watch it happen. The keyframe (I-frame) looks horrible because it's got nowhere near enough bandwidth to represent the image properly. Then the P-frames bring it successively up to snuff, adding more and more detail, until - smack, 15 frames later, the cycle starts all over again with a lousy I-frame.

Whereas if the I-frame was a good-quality image in the first place, then the P-frames would have very little work to do to maintain that quality image throughout the GoP.
 
I'll post a clip. No doubt in my mind that it's keyframe pulsing.


Totally disagree. Of course a P-frame can be much higher quality than the I-frame. The P-frame is not targeted to match the I-frame, it's targeted to match the raw uncompressed frame that's fed to the coder. A P-frame looks at the changes between the previous frame, and the target frame. If the target frame is incredibly complex, and the previous frame is mushy, the P-frame will use its bandwidth to encode the changes between the mushy frame and the detailed reference frame.

Each frame in the GoP does not care what the prior frame looks like, it instead cares what the target frame (the uncompressed image) looks like, and makes its best effort to look like that target frame. So yes, cyclic codec pulsing is very possible if the I-frame does a poor job on the first frame in the GoP.

In the clip I will post, you can watch it happen. The keyframe (I-frame) looks horrible because it's got nowhere near enough bandwidth to represent the image properly. Then the P-frames bring it successively up to snuff, adding more and more detail, until - smack, 15 frames later, the cycle starts all over again with a lousy I-frame.

Whereas if the I-frame was a good-quality image in the first place, then the P-frames would have very little work to do to maintain that quality image throughout the GoP.

I totally agree with Barry that we are talking keyframe pulsing here. I am seeing exact the same thing in my footage. Bang, everything is gone, then successively getting better, and bang, everything is gone again. But we are talking about details here which were previously simply not resolved. Hence, its like "complaining" on a very high level ;-)
 
Totally disagree. Of course a P-frame can be much higher quality than the I-frame.

Usually not higher quality that the I-frame it's using as reference, unless you have a very low I:P ratio. You said earlier the P-frame directly following the "mushy" I-frame improves on it. This usually isn't the case.


The P-frame is not targeted to match the I-frame, it's targeted to match the raw uncompressed
frame that's fed to the coder.

But you're looking at the GH1 encode, not the uncompressed frame fed to the encoder. You're examining and decoding the GH1 stream, not the preceding generation of video; so the P-frame is matched to the I-frame. Farther on in the GOP, each P-frame is matched to the preceding P-frame if reference frames=1 . The 1st P-frame is matched to the I-frame. In the streams I have seen , this firmware uses 1 reference frame. This means that P-frame can only use that IDR-frame or preceding P-frame as reference. (It cannot use 2 P-frames away for example).

A P-frame looks at the changes between the previous frame, and the target frame. If the target frame is incredibly complex, and the previous frame is mushy, the P-frame will use its bandwidth to encode the changes between the mushy frame and the detailed reference frame.

Yes, this is true, you need more bandwidth to code the residual. (i.e need more bitrate for P-frames if there is higher scene complexity). It's a balancing act, between I:P and P:B ratios (P:B doesn't apply here). If you take too
much away from the I-frame, the keyframe looks poor, but the following frames in the GOP will also look poor if they do not have enough bitrate. If you use too much for the I-frame , the video deteriorate as you proceed in the GOP, as the P-frames do not have enough bitrate

Each frame in the GoP does not care what the prior frame looks like, it instead cares what the
target frame (the uncompressed image) looks like, and makes its best effort to look like that target
frame. So yes, cyclic codec pulsing is very possible if the I-frame does a poor job on the first frame
in the GoP.

But it does care what the prior frame looks like. It uses the prior frame as a reference, and it codes the residual between that reference and what is predicted; otherwise every frame would be called an I-frame and not a P-frame:) If you had higher # of reference frames, it could use P-frames as reference farther away and coding would be more efficient. If you had B-frames, you could use them in both directions and this would be even more efficient in theory.

In the clip I will post, you can watch it happen. The keyframe (I-frame) looks horrible because it's
got nowhere near enough bandwidth to represent the image properly. Then the P-frames bring it
successively up to snuff, adding more and more detail, until - smack, 15 frames later, the cycle
starts all over again with a lousy I-frame.
Whereas if the I-frame was a good-quality image in the first place, then the P-frames would have
very little work to do to maintain that quality image throughout the GoP.

I see what you're saying. The "textbook" case of keyframe pulsing is high I-frame allocation, and low P-frame allocation. The quality slowly deteriorates proceeding along the gop until the next high quality I-frame and it "pops" into focus and high quality. This is the far more common scenario, you even see it in retail blu-ray , online on vimeo, youtube etc... I apologize for the mixup, I should have read your description more clearly
 
Last edited:
I see what you're saying. The "textbook" case of keyframe pulsing is high I-frame allocation, and low P-frame allocation. The quality slowly deteriorates proceeding along the gop until the next high quality I-frame and it "pops" into focus and high quality. This is the far more common scenario, you even see it in retail blu-ray , online on vimeo, youtube etc... I apologize for the mixup, I should have read your description more clearly
Yeah, that's the exact opposite of what we're encountering. Quality is awful for a frame and then slowly improves, until it "pops" back mushy on the next I-frame.

That's why I say that for this case, what we really need is more bandwidth allocated to the I-frame. This may not be relevant to long-GoP in general, but in the specific case of the GH1 it seems to be exactly the medicine for the problem at hand. :)
 
Can you please still upload the clip Barry? I'd still be interested in seeing it

I know the initial few seconds can be problematic as you start recording, but does this occur as bad throughout even 5-10 seconds later?

I sold my GH1 a while back , but these recent developments have seriously piqued my interest!
 
I can't seem to find anything/anyone that speaks about the Lumix's HDMI out while recording. What's it doing? Full-res? Any menu-stuff on-screen?
Thanks.
 


Thanks for uploading that.

This stream has errors. There are frame repeats before the "truncated" I-frames in the previous GOP. Not all I-frames are low quality, there are "normal" looking GOP sequences, which do not have frame repeats.

In this example frame 11+12 are frame repeats (as are all the others before the "truncated" I-frames , 23+24, 35+36, etc...)

2you7lu.jpg


What settings did you use for this clip?

Do you have a longer example? I alluded to bad frames at the beginning of recordings - this is common and can occur even in other camcorders, and the unpatched GH1 - but not to this extent. It's not limited to the beginning or "warm up" period either - There is a bad I-frame even in the middle of the clip surrounded by "normal" I frames. (you can see it "shortened", in this graphical representation the height corresponds to the coded frame size, red frames are I-frames, blue are P-frames)

This stream also has buffer underrun errors. Normally this isn't a problem for PC, but some devices and hardware decoders may have issues decoding it.

This stream also has incorrect FPS at the bitstream level, may cause sync issues on longer compositions and in some NLE's

Also, it uses 22 slices/frame => can cause problems with some decoders which usually max out at 16
 
Last edited:
That's the only clip I have that has the pulsing in it. I don't remember the exact settings, but they were the "D" settings when those were first announced.
 
i experienced the pulsing when shooting 720p50 at frist 2-3 second which skip one of two frames. so i recall my setting and remember the shutter speed below 1/50. my cam is patched to "D" setting. then examed the pervious footages at both 1080p25 and 720p50 and found normal.

i am trying to duplicate the pulsing situation, any idea of it?
 
i've had this pulse since i first got my GH1 also with the GH13 always noticeble in 720 60p (high detail scenes ) all the way through clips. to be honest i haven't noticed it in FHD.
can it be fixed?
 
PTool 3.36 Updated without version change.

Changed status of Version Compare patch. It is advisable to use it instead of increments for GH1.
MJPEG Encoder patch removed from list, will be applied automatically upon MJPEG width and heigh change.
Video Buffer patch removed.
 
Great Tool adjustments. About to check MJPEG tomorrow.

Tester: so Video Buffer was doing nothing at all? When I tested, and reported back, anything over 40 would give my camera errors, but people are using 84 and having no issue.

Nice Avatar. Haha.
 
I did diferent GF1 low light tests with 1440 *1080 mjpeg and avchd and it seems the the mjpeg has no noise reduction applied. Can anyone confirm this ?
 
PTool 3.36 Updated without version change.

Changed status of Version Compare patch. It is advisable to use it instead of increments for GH1.
MJPEG Encoder patch removed from list, will be applied automatically upon MJPEG width and heigh change.
Video Buffer patch removed.

You need to upgdate the screenshot on the first page.

Thanks and good luck.
 
i've had this pulse since i first got my GH1 also with the GH13 always noticeble in 720 60p (high detail scenes ) all the way through clips. to be honest i haven't noticed it in FHD.
can it be fixed?

Is it definitely not just down to the playback?

I noticed it at first but later realised it was dependent on the software I used to play/convert the file. If you convert properly you don't see it.
 
Back
Top