Panasonic Removable Lens Four Thirds HD AVCHD Camera

Update on the 5DM2 missed pixel issue.

Here is the evidence:


  1. To record everything, the camera would have to process 18Mp/frame x 30 frames/sec or 540Mp/s. People in the know say that isn't possible. For still images, the 1DsM3 does 5 21Mp frames/sec, or 105Mb/s.
  2. There is a vertical moire pattern in some 5DM2 footage. Here is an example where the raw footage can be downloaded: http://www.vimeo.com/2738955. That would be an example of reduced vertical resolution. Notice it in the roof of the house.
1236715066.jpg


The consensus seemed to be that the Canon process every third row for 6Mp/frame or 180 Mp/s. That is probably within what most seem to think is a reasonable bandwidth. The Moire pattern results in having a low-pass optical filter for 78 lp/mm while delivering 1/3rd that. Basically, the optical filter passes everything vertically.

Further evidence how good the recording must be.

For comparison, the GH-1 can process 3 12Mp frames/sec in burst mode, or 36Mp/s. Casio's high speed Exlim can record 60 6Mp frames sec or 360 Mp/s. So the 210Mp/s seems possible.


How does this affect the Panaonic? The GH-1 has a 12.1 Mp 4000x3000 sensor. for 16:9 that would be 4000x2250. For 1080p24 that would be 9Mb/frame x 24 frames/sec, or 216 Mp/s or slightly more than what users are guessing the 5DM2 is doing.

However, what about 720p60 mode? That would have to be 540Mp/s, which everyone thinks is rather unrealistic. Does it use a 2560x1440 section of the frame in the center of the sensor? That would be 221Mb/s however it wouldn't work with the optics. The resulting 11mm wide sensor would have no wide angle at all and stills taken during shooting wouldn't match.

If they dropped every third row (to maintain the bayer pattern) it would be subject to the same moire pattern as the Canon.

However, until it was pointed out I hadn't noticed the moire pattern in an clips and still find it to be fairly rare (really, only the clips someone pointed it out it).
 
Last edited:
I am glad the link may have been useful. It does seem indeed you have the right skills in order to proceed with your findings. :dankk2: for sharing.
 
DavidNJ said:
Update on the 5DM2 missed pixel issue.

Here is the evidence:


  1. To record everything, the camera would have to process 18Mp/frame x 30 frames/sec or 540Mp/s. People in the know say that isn't possible. For still images, the 1DsM3 does 5 21Mp frames/sec, or 105Mb/s.
  2. There is a vertical moire pattern in some 5DM2 footage. Here is an example where the raw footage can be downloaded: http://www.vimeo.com/2738955. That would be an example of reduced vertical resolution. Notice it in the roof of the house.
1236715066.jpg


The consensus seemed to be that the Canon process every third row for 6Mp/frame or 180 Mp/s. That is probably within what most seem to think is a reasonable bandwidth. The Moire pattern results in having a low-pass optical filter for 78 lp/mm while delivering 1/3rd that. Basically, the optical filter passes everything vertically.

Further evidence how good the recording must be.

For comparison, the GH-1 can process 3 12Mp frames/sec in burst mode, or 36Mp/s. Casio's high speed Exlim can record 60 6Mp frames sec or 360 Mp/s. So the 210Mp/s seems possible.


How does this affect the Panaonic? The GH-1 has a 12.1 Mp 4000x3000 sensor. for 16:9 that would be 4000x2250. For 1080p24 that would be 9Mb/frame x 24 frames/sec, or 216 Mp/s or slightly more than what users are guessing the 5DM2 is doing.

However, what about 720p60 mode? That would have to be 540Mp/s, which everyone thinks is rather unrealistic. Does it use a 2560x1440 section of the frame in the center of the sensor? That would be 221Mb/s however it wouldn't work with the optics. The resulting 11mm wide sensor would have no wide angle at all and stills taken during shooting wouldn't match.

If they dropped every third row (to maintain the bayer pattern) it would be subject to the same moire pattern as the Canon.

However, until it was pointed out I hadn't noticed the moire pattern in an clips and still find it to be fairly rare (really, only the clips someone pointed it out it).
Click to expand...

David, are you saying that Liza's information was correct on how the MKII gets its video in the end or is it the opposite?

Trying to make sense of this post altogether. Forgive my ignorance.
 
Some know a lot more...

More info

The Venus Engine HD also supports an extensive range of functions, including HDMI output.
Click to expand...
That may be the answer to another VERY important question.

And

05_img05.gif
Click to expand...

The OIS has a panning mode. Well at least on the G1 lens. The GH-1 isn't specific.

The Venus Engine HD incorporates
two CPUs. This not only boosts image
processing speed, it also helps achieve an
exceptional noise-reduction performance.
Also, the two CPUs of the Venus Engine
HD allow long-time movie recording in
AVCHD while maintaining low power consumption.
Click to expand...

another new technology makes it possible to read 4 channels of data simultaneously, which allows the DMC-GH1 to deliver Full-time Live View images at 60 frames per second, while faithfully reproducing high-resolution images with fine detail and rich gradation.
Click to expand...

Taken together the sensor allows 4 channels to be read at once. The two CPUs allow 'low power consumption'. The Canon's overheating problem may come from running the processor at a fast speed (power, which implies heat dissipation, increases linearly with processor speed and as the square of voltage).

The high required bandwidth may be the result of multiple parallel processing streams...540Mp/s?

P.S.
A three sensor 1080p camera (Sony EX1/EX3, Panasonic HPX300) would process 2Mp/frame * 30 frames/sec * 3 sensors = 180Mb/sec.
 
Last edited:
What do you mean? LoL. I don't care who was right or wrong, the information is astounding on both ends and keeps me interested. I ask because Liza also mentioned that the GH1 may really shine at 17mbps because it doesn't scale like the Canon does. So if the info presented by Liza is right, that gives me even more hope for this camera.

The OIS panning mode may be why we don't see a lot of jello. But if it's a lens feature and not a camera feature then those of us planning to use other lenses will get just as much jello as before.

HDMI output: please oh gahd please. Seriously, this is getting exciting!
 
Kholi said:
David, are you saying that Liza's information was correct on how the MKII gets its video in the end or is it the opposite?

Trying to make sense of this post altogether. Forgive my ignorance.
Click to expand...


Lisa may have been correct concept but not in detail. The information isn't definitive, however, this is the source information that would have supported the proposition. Note, the detail is different. They hypothesis among users who encountered it is that it is scanning 5616x1080, alternating every third row.

The GH-1 sensor has about half the number of pixels (in 16x9 mode), so it is starting closer. And p24 drops that to 40%. 720p60 would bring the total right where the Canon would be if every photo site was scanned. So some questions remain.


Note, ABC, Fox, ESPN all broadcast 720p60, do consider it lower quality than CBS, NBC, or HBO?
 
David, I have not watched the video since yesterday but from what I remember I did not notice the skew while he was in 60p mode.....only in 24p. Do you think what you said here has anything to do with that? Or was I just seeing things?

The GH-1 sensor has about half the number of pixels (in 16x9 mode), so it is starting closer. And p24 drops that to 40%. 720p60 would bring the total right where the Canon would be if every photo site was scanned.
Click to expand...
 
Not sure...which video...however, the hypothesis is that it would occur on 720p60 and not 1080p24. From a capture standpoint, one is reading the sensor 2.5x the other.
 
Kholi said:
David, are you saying that Liza's information was correct on how the MKII gets its video in the end or is it the opposite?

Trying to make sense of this post altogether.
Click to expand...

Kholi said:
What do you mean? LoL. I don't care who was right or wrong, the information is astounding on both ends and keeps me interested. I ask because Liza also mentioned that the GH1 may really shine at 17mbps because it doesn't scale like the Canon does. So if the info presented by Liza is right, that gives me even more hope for this camera.
Click to expand...
DavidNJ said:
Lisa may have been correct concept but not in detail. The information isn't definitive, however, this is the source information that would have supported the proposition. Note, the detail is different. They hypothesis among users who encountered it is that it is scanning 5616x1080, alternating every third row.
Click to expand...
The 75% are not accurate. Not exactly the same as the DOF information -- where it is quite the opposite, actually. But close. On those 75%, they become 66% as matter of fact. What it doesn't mean any prevail of a 2x smaller sensor size, whether in terms of depth of field or even whether light sensitivity. Oddly or maybe not so oddly, the quality is not only measured in pixels. The case of its size. In small formats, the more pixels the smaller they are, with no absence of compromisses. And for instance, as we all know, bringing the inevitable noise or not necessarily more than another level of difficulty layers of sensitive response. The case of L lens(es) of large image circle already referred is not effectively non-sense at all, as well. As any real 35mm shooter can notice and knows it very well. Or yet, if not, why so much interest on fast lenses then after all?

In two words, everything is resumed to 'size' and 'light'. Both count. Quality doesn't necessarily mean number.
 
A couple of things...

1) Pixels are not a good indication of picture quality. There is quite a few SD video work that would be better than much HD. What you are interested in is contrast often measured as MTF. At low frequencies it is displays picture edge sharpness and where the in camera sharpening affects damage the picture. In mid frequencies it adds texture. HD lets you capture that texture (usually lost in SD) IF the lens and in camera processing can deal with it. Small sensor, small lens HD cameras rarely have much in the way of texture. Many first generation HD systems also lacked texture. The new systems have more. The 5DM2 has tons of texture. Really high resolution, near the Nyquist limit is useful for aerial reconnaissance.

1236734808.jpg

​
2) DOF depends on the aperture, size of the sensor, and focal length of the lens...whether it has 10,000 photo sites or 10 billion. This is a good explanation: http://www.cambridgeincolour.com/tutorials/depth-of-field.htm


3) Low light sensitivity is a function of the amount of signal captured. That is a fuction of photo cell design and quality and its area. The high count is often better because the inter photo site gaps are smaller.

CanonEOS-02.jpg

​
4) Noise, is a function of the quality of the electronic processing. The sensor and the subsequent processing. Obviously, the signal to the sensor, the light, is affected by the glass, the Bayer filter, the microlens, etc. The A/D conversion has to read an electrical signal from the sensor and convert it to a digital value. Gain means it is reading a very small signal for the complete dynamic range and expanding it. Now consistancy between each sensor and the circuits reading their voltage or current levels because the issue. Since small sensors have small cells it makes it harder. The GH-1 has twice the pixel density as the 5D Mark II, so the photo site to photo site varition may be larger. However, although good (and very expensive) 35mm glass is available, often APC and presumably similar sized 4/3rds glass often has less distortion. Overall, the 4/3rds at similar resolution may be close overall.
 
Last edited:
DavidNJ said:
A couple of things...

1) Pixels are not a good indication of picture quality. There is quite a few SD video work that would be better than much HD. What you are interested in is contrast often measured as MTF. At low frequencies it is displays picture edge sharpness and where the in camera sharpening affects damage the picture. In mid frequencies it adds texture. HD lets you capture that texture (usually lost in SD) IF the lens and in camera processing can deal with it. Small sensor, small lens HD cameras rarely have much in the way of texture. Many first generation HD systems also lacked texture. The new systems have more. The 5DM2 has tons of texture. Really high resolution (...)
Click to expand...
Can't agree more. Your 2nd part justifies the reason of the counterculture of a 1st assessment politically incorrect.

DavidNJ said:
2) DOF depends on the aperture, size of the sensor, and focal length of the lens...whether it has 10,000 photo sites or 10 billion. This is a good explanation: http://www.cambridgeincolour.com/tutorials/depth-of-field.htm
Click to expand...
As quoted, size where sensor and lens may shine. Therefore, the larger, the more number of 'layers' aka quality of bokeh. Up to a certain kind of limit or balance. It's often to listen from ACs: «the Vistavision (circa FF35mm format) is cruel».

DavidNJ said:
3) Low light sensitivity is a function of the amount of signal captured. That is a fuction of photo cell design and quality and its area. The high count is often better because the inter photo site gaps are smaller.

CanonEOS-02.jpg


4) Noise, is a function of the quality of the electronic processing. The sensor and the subsequent processing. Obviously, the signal to the sensor, the light, is affected by the glass, the Bayer filter, the microlens, etc. The A/D conversion has to read an electrical signal from the sensor and convert it to a digital value. Gain means it is reading a very small signal for the complete dynamic range and expanding it. Now consistancy between each sensor and the circuits reading their voltage or current levels because the issue. Since small sensors have small cells it makes it harder. The GH-1 has twice the pixel density as the 5D Mark II, so the photo site to photo site varition may be larger. However, although good (and very expensive) 35mm glass is available, often APC and presumably similar sized 4/3rds glass often has less distortion. Overall, the 4/3rds at similar resolution may be close overall.
Click to expand...
The 35mm lenses will equal in distortion available with their sweet centre. However, there's less surface, more pixels for fighting there for less light available per each. Again there's a balance. In your finest example (5D vs. 5D Mark II), there was territory for exploring. But the sweet limit is close to be beaten. Such as it has occurred with the small formats more than ever, 1st than with any larger format -- 2nd round. The time has come. Michael Reichmann from Luminous Landscape, for example, has disadvised the new Nikon D3x. Reason? Why pay twice what you can have for mid-price? The 3.5MP don't worth the (slight) difference, in his opinion. We can compare apples to apples but we should not do with oranges. Or lemons.

IMHO, a small format is convenient for reasons of portability and handheld acquisition where others do not dare to enter, DOF for docs, guerilla style or 3D/stereoscope production for instance. Otherwise, it smells less than cinematic. I doubt GH1 may surpass it. Will do comparing with others.



EDIT -- I am not so enthusiastic with the new industry standard as Bruckheimer. Aesthetic reasons. But I will produce a 3D feature next year. :-)
 
Last edited:
Back when I was in hardware engineering, prototypes would run at a reduced clock rate while we worked out problems. I don't know if the pre-production models Panasonic were showing at PMA are doing this (a possible cause of the banding) or if they are running full stop. If they are running at a reduced speed, that could well mean that the skew in the final products is going to be less than we are seeing.

I was hoping for no skew, and I'm concerned about the skew we're seeing, but it looks like it will be manageable.

Regarding the video pipeline, David's confirmed what I was saying. The %75 figure came from assuming the canon was using every other line instead of every third line. IF you throw out half the lines vertically and half the lines horizontally you are reducing the pixel count by %75. Since its every third line, if that's in both directions the reduction is greater. However it may not be in both directions, and it may be more complex than that (Eg: they have some flexibility given how large their sensor is in MP.) But I guess every third line could work, that just seemed high to me.

On the GH1 it seems reasonable that quad read channels from the sensor and dual CPUS allows the GH1 to have higher bandwidth, and these numbers indicate the bandwidth expectations are not that high. Lets hope Panasonic has tried to take a no-compromises approach with this design.
 
Last edited:
LizaWitz said:
Lets hope Panasonic has tried to take a no-compromises approach with this design.
Click to expand...

The GH 1 is basically a video camera developed on the G1 platform. Almost all the differences it has compared to the G1 are meant to produce good video. It is NOT a DSLR or an EVIL camera with an additional video function. Said this I'm very optimistic they did the best job they could to create a functional video camera with high image quality.
 
It is still the consumer photo division vs. the pro video department. They did somethings but there are lots of things we look for on a video camera (zebra, color matching adjustments, knee control, black stretch, headphone jack with output control, etc.) that aren't there.

The key is the quality...that is what gave interest to the D90 and made the 5DM2 a paradigm changing event. People live with lots of aggravation to use a 5DM2; their result is an average image that often makes the best EX1/EX3 image look like a mundane.

Hopefully the GH-1 (with HDMI output and manual controls) will let us equal or exceed the 5DM2. However, two years from now it will also be obsolete.
 
You must log in or register to reply here.
Back
Top