GH1 Not as I had hoped

[Steve Mullen]

Originally Posted by M. Gene Hoffman

The WHY behind any of this doesn't matter if the end result is that the footage sucks for many kinds of shots.
-M

YOU tossed in this comment, "but this just looks like Panasonic being lazy with their implementation."

If you are going to comment in WHY something sucks, be prepared to be called on the accuracy of your comment. And, don't weasel out with a doesn't matter why -- it just sucks.

If it doesn't matter, then why did you feel the need to explain why it sucks?

--------

PS: It really doesn't matter:

1) If a Canon camcorder shoots better video. It's still a tiny little no VF camcorder. Or, it has a VF, but uses tape. These tiny cams from Canon, Sony, and JVC are a dime a dozen. And, they don't shoot 720p60.

2) If the 5D shoots better video. It's also 2X more expensive. And, at 2X more there's a camera that shoots even better video. And, it better shoot 2X+ better video.

Someone seeing the Rebel at Costco for under $700 isn't going to think, "gee maybe I should spend 2X and get a bit better HD." That's not how consumers buy. (It's why Sony sells fewer than Vizeo.)

[M. Gene Hoffman]

Steve,
You are beating around the bush, which is why I chose to ignore your argument. I don't want to argue technical stuff- I am as comfortable as anyone doing so, but it has nothing to do with the point of the whole thread.

The point is that the footage doesn't look good for many types of shots. And it would with a better codec implementation.

Making a breakthrough camera that exceeds in all areas except for the last one that matters comes across as lazy. Everybody else has a good AVCHD codec implementation. I am aware that Panasonic does too in their higher end stuff. The whole point is that they didn't use it here, and for whatever reason, it comes across as stupid or lazy, because the rest of the camera is superb.

-M

[Steve Mullen]

Originally Posted by Steev Dinkins

The question remains. How do you convert 60p to 24p? And I'm not talking about for slow motion (conform 60p to 24p).

The point of 24fps is NOT the frame-rate, but the temporal sampling TIME that is so long (1/24th second) that motion can NOT be captured accurately. The Nyquist frequency of sampling is so low, motion aliases and we see it as judder. Or, more simply put, we see unsmooth motion.

When p60 is dropped into a p24 timeline the NLE treats the 60 frames as 12 groups of 5 frames each: A, B, C, D, and E. The time between each capture is 1/60th second.

It now ignores all but frames A and C. The time between A and C is 1/30th second while the time between C and A is 1/20th second. The average inter-sample time is 1/25th second -- low enough to have all the qualities of "film". And, a perfect match for a 1/50th shutter-speed. You now have a timeline with the TEMPORAL nature of a video shot at 24p.

----------

Of course, you cannot view it at 24fps. Your LCD display is running at 60Hz. So when you hit Play, frame A is presented twice and frame C is presented three times. Since there are 12 As and 12 Cs -- every second the display shows, 24 As and 36 Cs for a total of 60 frames. This is 2-3 pulldown applied to progressive video.

In this 60 frames, the temporal sampling time remains ONLY about 1/24th second. The motion blur for each frame is still about 1/48th second.

---------

You Master to 720p60 using 2-3 pulldown. The Master goes to BD at 720p60. The HDMI connection feeds 720p60 to HDTVs. The display shows each frame once, twice, four, or eight times. Nevertheless, within these 60/120/240/480 frames, the temporal sampling time remains ONLY about 1/24th second. The motion blur for each frame is still about 1/48th second. It still looks like film -- unless you enable frame interpolation in the display. If you do that, the display will tween frames to increase motion accuracy. So, turn it off!

-----------

The key is stop thinking of film and it's historical 24fps. Think in terms of temporal sampling ACCURACY. The supposed magic of film is it is has very low ability to capture motion accurately -- about every 1/24th second. (I say "about" because until modern days, film cameras never captured frames EXACTLY 24.0000 second apart.)

So there's your modern workflow that starts with 720p60. Yet, to the viewer, it looks like you shot video at p24.

PS: Yes you have wasted some bits. But, not the number you think. Changing frames every 1/24th second -- when there is motion -- forces P and B frames to be very large. When captures are made every 1/60th second, P and B frames are tiny. Therefore, ignoring lot's of P and B frames with p60 isn't wasting lots of bits.

How about ignored I-frames? If I remember correctly, AVCHD has a 15-frame GOP. So in 60 frames there are 4 GOPS:

Abcde abcde abcde Abcde abcde abcde Abcde abcde abcde Abcde abcde abcde

Looking only at the 24 frames that are not ignored:

Ac ac ac Ac ac ac Ac ac ac Ac ac ac

There are 4 I-frames every 24-frames!

What if you shot true 24p with a 15-frame GOP?

Abcdefghijklmno Pqrstuvw

Only ABOUT 2 I-frames every 24-frames. Fewer I-frames every second.

Bottom-line -- shooting 720p60 you waste mostly P- and B-frames, but you preserve more I-frames.

PS: of course, for segments you can still create slo-mo from the 60p source.

[Steev Dinkins]

I'm seeing this derailing off topic.

I've started a new thread to outline my findings on 60p to 24p conversion here:
http://www.dvinfo.net/conf/panasonic...onversion.html

[Steve Mullen]

Originally Posted by M. Gene Hoffman

The i frames are SUPER obvious on the panasonic,

There is popular belief that there is a visual difference between I-frames and P- and B-frames. This comes from several miss-understandings of how long GOP works.

1) The point of any compression system is to keep quality constant -- although with CBR that can't totally happen because the bit-rate is fixed, so quality is allowed to vary. With VBR, the quality can indeed be constant.

The reason the P- and B-frames are smaller is because there is less information to compress since they only carry changing information. With a constant quality goal -- naturally the amount of recorded data will be less for these frames. The fewer bits in no way implies lower visual quality. Likewise, the big I frame does not look visibly better. All frames after decoding have the same visual quality.

2) The other reason folks believe the further from the I frame -- the lower the visual quality -- is because they think P- and B-frame are predicted from the I-frame. Hence, the prediction must get less accurate going through a GOP. Not how it works.

Each P- and B- frame contain two types of information. One type is indeed the motion vectors that attempt to accurately place pixels in the next frame. Obviously, there is a chance of errors. So, after creating these vectors, they are used to generate a "next" frame.

Then, the errors between the generated and actual next frame are computed. This DIFFERENCE information is the second type of information.

Both types then undergo bit-reduction.

During decoding, the motion vectors generate a potential next frame and then the DIFFERENCE information is applied to it. The corrected, now very accurate frame, is output. When you combine this process with the dual direction predictions of B-frames, there are no fluctuations of visual quality within a GOP.

---------

But, what happens at scene changes? Depends on the encoding rules. If the GOP can't be shortened and the system is using CBR, the frames following a scene change until the next I-frame will have compression artifacts. However, with a VBR system plus the ability to start a new GOP, plus h.264's artifact filtering function -- the inserted I-frame is far more likely to handle the scene change without artifacts.

You'll note that reports say when the exposure fluctuates rapidly, quality drops. That's because each fluctuation is like a new scene. Long GOP systems can't keep shorting GOP.

----------

1920x1080 has 2X more pixels to compress than 1280x720. 720p60 has 2X fewer frames than 108060i. They balance each other out -- although progressive frames compress more efficently which is why ATSC uses 18Mbps for 1080i60 and only about 14Mbps for 720p60. (This allows carrying an extra SD channel.)

Were 1080p30 being carried, then it too would also be fine at about 14Mbps. And, 1080p24 could use as little as 11Mbps. So were a camera a broadcast station, they would likely squeeze 1080p24 to 10-12Mbps. But, a camera isn't a broadcast station. There is no reason to limit data rate for 1080p24 and market it as 17Mbps. Once marketed at 17Mbps they can't claim 2X longer record times for p24.

So if indeed 1080p24 is using a 2X lower bit-rate -- then someone set the bit-rate reduction ratio for 1080p24 and 720p60 to the same value -- based upon a limit of 17Mbps. Once this was done, 1080p24 would logically have about a 2X lower bit-rate. That's simple math.

[Wacharapong Chiowanich]

Originally Posted by M. Gene Hoffman

it comes across as stupid or lazy, because the rest of the camera is superb.

-M

Think about it. If a video camera manufacturer with the resources at the level of Panasonic has overcome so many technological hurdles in bringing about good video images to the mass and yet seemingly fallen at the very last hurdle before the images out of a camera such as the GH1 can be seen on the screens, this could be purely intentional, not them being stupid or lazy. Imagine the GH1 being able to shoot superlative 1080/24p images at 24 Mbps or at 17 Mbps at a quality comparable to that of the Canon HF10. Think about how many potential buyers of the prosumer AVC cams or the consumer AVCHD series Panasonic would lose to the new GH1. As it already is, most target buyers of this cam would never notice the shortcomings like motion artifacts in some demanding situations, for instance.

I think they have thought this through and done their homework with respect to the marketing of this camera.

Wacharapong

[Joe Kowalski]

Originally Posted by Wacharapong Chiowanich

Imagine the GH1 being able to shoot superlative 1080/24p images at 24 Mbps or at 17 Mbps at a quality comparable to that of the Canon HF10. Think about how many potential buyers of the prosumer AVC cams or the consumer AVCHD series Panasonic would lose to the new GH1.

Wacharapong,

Allow me to direct you to the TM300:
Panasonic HDC-TM300 HD Camcorder | Wired.com Product Reviews

It's $1300.
It has a 3 chip design.
It has a Leica F1.8-2.8 44.9-539mm lens -- less wide than the GH1, but longer tele, and MUCH faster. I'm guessing it has a zoom rocker, too.
It has 32GB of built-in storage.
It has 5.1 surround sound microphones -- sounds like a gimmick, but people love gimmicks.
It even takes 10 megapixel shots, nearly the resolution of the GH1's stills.

The best feature: "The auto-focus tracking feature is straight out of the future: Tap a subject on the LCD, and the camera automatically keeps it in focus as your target moves about the shot."

Some of that sounds pretty nice. I'm much more interested in the GH1, but I think for the average consumer, the TM300 is going to look much more appealing.

[Steve Mullen]

Originally Posted by Joe Kowalski

Wacharapong,

Allow me to direct you to the TM300:

Reviews are up and this looks to be a great camcorder. But, it lacks not only 720p60 but 1080p30.

However, wait until August and spend 2X more and get the really great Panasonic HVM40.

Think of a true PRO version of the TM300 with 720p60 and 720p30.

This will likely blow away the GH1, but then at 2X more money it should. But, for those who also want to shoot stills -- the GH1 is a nice package.

PS: the HVM40 uses 24Mbps AVCCAM. Nice, but good luck editing it.

[Robin Lobel]

Steve: Where did you heard about "Panasonic HVM40" ? I cannot find a single reference to it on the internet. Plus, would it have a sensor as big as the GH1 ? That's the only way it can blow the GH1 away...

[Steve Mullen]

Originally Posted by Robin Lobel

Steve: Where did you heard about "Panasonic HVM40" ? I cannot find a single reference to it on the internet. Plus, would it have a sensor as big as the GH1 ? That's the only way it can blow the GH1 away...

Sorry. HMC40.

How about a waveform monitor on the LCD? How about full audio controls? How about a power zoom? And, most import, built-in ND filters. Plus, all frame-rates and frame sizes. Oh, and let's not forget 24Mbps.

Sensor size is not such a big plus as narrow DOF only increases the likelihood of out of focus shots.

[Robert Morane]

Steve, do you know anything about the lense? Focal lenght and diameter. Thanks

[Steve Mullen]

Originally Posted by Robert Morane

Steve, do you know anything about the lense? Focal lenght and diameter. Thanks

Announced and shown at NAB. Should be all over the web and at Pana site.

[Cole McDonald]

Originally Posted by Ian G. Thompson

You're right...24p is no limitation. There is definitely an obvious difference between 24p and 30p and 60p. I don't know what this animosity towards 24p is. I've been hearing a lot of this talk m

I'm curious as to how much of the film look is delivered via the actual frame rate and how much the standard shutter speeds associated with those frame rates.

I'd like to see a sample of 24p 1/48 next to some 30p 1/48 to see how well it holds onto the "Film Look" From a purely physics stand point, the 1/48 shutter seems more relevant to me than the actual frame rate as that defines how much motion blur is presented to the viewer. For that matter, Is it more filmic to shoot 30p 1/30 if 1/48 isn't available?

Would someone be interested in testing these? Shooting in 30p for 60i delivery would be much easier and cheaper to post than the "is everything right for working with 24p" game we need to play to work with 24p stored in a 60i stream. But there are too many hoops to jump through to work with it. I know the hoops and I still forget every once in a while. PITA!

[Steve Mullen]

Originally Posted by Cole McDonald

I'd like to see a sample of 24p 1/48 next to some 30p 1/48 to see how well it holds onto the "Film Look" From a purely physics stand point, the 1/48 shutter seems more relevant to me than the actual frame rate as that defines how much motion blur is presented to the viewer. For that matter, Is it more filmic to shoot 30p 1/30 if 1/48 isn't available?

It is a combination of the interval between captures (1/30th vs 1/24th second) and shutter-speed (1/60th vs 1/48th second).

30p captures motion slightly better -- which is oddly what the film folks like. 1/48th offers slightly more blur -- which is needed to help make 24p judder be less severe.

I completely agree that 30p makes so much more sense. And, I really doubt an audience can tell the difference. However, when I use 30p -- I use 1/45th (about 213-degrees if I remember right. 1/30th is so blurry it reduces effective resolution. (Of course, that makes some think video looks like film -- which is it doesn't.) And, 1/60th has almost no motion blur.

[Robin Lobel]

Originally Posted by Steve Mullen

Sorry. HMC40.

How about a waveform monitor on the LCD? How about full audio controls? How about a power zoom? And, most import, built-in ND filters. Plus, all frame-rates and frame sizes. Oh, and let's not forget 24Mbps.

Sensor size is not such a big plus as narrow DOF only increases the likelihood of out of focus shots.

Erm. If people talks about GH1, it's because it's the first videocamera with full manual control AND the big sensor. If it would not have that big sensor, there would be no buzz at all.
So you can't compare HMC40(1/4" sensor) vs GH1, it's a completely different market.

Finally, GH1's 1080p compression looks pretty good for most situation:
http://www.vimeo.com/4582107
It only breaks at the end (last shot with action), which would be handled properly using the 720p60 mode.