AF100 shows more aliasing and less resolution than Panasonic GH2 ?

To go along with the AF100 sitting here, Ive been wanting and talking about dumping the 5dmkii for a gh2 - but I just cant get over the line on the big fat 0 1080/25P (Im Aus, Aus is a PAL country). GH2 seems to solve many 5dmkII issues as well as being smaller.
 
Barry_Green said:
(and I'm not saying the German site is wrong, yet... I'm just acknowledging that there's a whole lot more going on than can be explained by a black & white chart. I can easily get aliasing on the AF100 and GH2 on a b&w chart, but cannot make it happen in real life on bricks, which is the exact opposite of my experience with the GH1 and 7D. I don't know if there's some sort of sophisticated processing going on that detects and compensates for so-called real-world images, maybe there's something in there that uses color in some sort of way... I don't know. It's odd. But on B&W I can get the AF100 to alias like shown above, yet in a color chart like The Wringer, it looks gorgeous out to 800 TVL).
Click to expand...

Looks like the GH2 is exhibiting the HDMI chroma issue documented in the GH2 forum, where there is some weird jagged chroma in the HDMI apparently somehow due to some odd 60i (not 59.97i) conversion happening before HDMI output, all I could gather form that thread is that it is fixable in post if the footage is interpreted with a script someone came up with, but it doesnt show up in internally recorded footage anyway. Besides that, I'm surprised the AF100 and GH2 look so similar on the "wringer" considering slashcam (the german site) had an MTF curve showing much lower chroma resolution for the GH2 compared to the AF100, despite its apparently higher luma resolution.

Definitely odd that the wringer looks so clean on the AF100 when there is (some) aliasing on the similar black and white chart. Maybe something odd going on with the camera's internal RGB->YUV conversion?
 
Noel, I still don't get their motivation on why they left that basic fundamental feature out. It makes not sense.
 
Barry_Green said:
And I'll get there... I'll get to it... I just feel like trash and staying upright is not the easiest nor wisest course of action right now. :thumbsup:
Click to expand...

Get well soon Barry. It sounds like you need the rest. I am going to try to do the tests you mentioned with Rick and the AF100 tomorrow. We will try the resolution chart and a color resolution chart I also have to confirm the German site's results. Then we can shoot some test shots of the brick buildings to see if they still hold up.
 
As the "initiator" of this thread I am delighted that you guys are addressing this issue with such ferveur.

There may be a possible reason for the disconcerting and apparent incoherence between the AF100 moiré showing up in the German test vs the clean picture in Barry's test, when dealing with these strange test results: it could be caused by the already noticed imbalance of the chroma channel, already linked with the weird chroma cut-off producing non-white clip off highlights. Perhaps the cam is aliasing differently for each chroma channel?

Is the AF100 a sloppy product , meaning that it is hastily and poorly calibrated ? Or is the GH2 artificially "face-lifted" in software?

Agree that a far more rigourous testing involving the chroma channels is needed. I hope that Barry recovers soon to decrypt this Da Vinci code...
 
sloppy is an unfair description. the moire is minimal. not even noticeable at all in any shots. compared to a dslr where moire happens anywhere.
 
I wonder if what we are seeing is differences in the filter pattern possibly. As Panasonic has not said whether the AF100 uses a Bayer pattern or not, given that we are seeing different results in different contexts of the images, it seems reasonable to suggest that this is not simply an apples to apples test. The follow on question to me is that if the AF100 is not using a Bayer pattern, is it to use other methods that can be employed to increase DR? Reduce noise?
 
Further thought based on Rick's reference to the Bayer pattern:

Could it be the case that the Bayer filter and the OLP filter are combined in one single filter, with the result that the "low-passed" resolution is different for each chroma channel when hitting the photosites ?
 
Last edited:
danap said:
Further thought based on Rick's reference to the Bayer pattern:

Could be the case that the Bayer filter and the OLP filter are combined in one single filter, with the result that the "low-passed" resolution is different for each chroma channel when hitting the photosites ?
Click to expand...

This wouldn't work. The reason being, the image material within the color filter boundary actually needs to move to other pixel locations, diffused out, not within the pixel itself. For instance, if you have a blue dot over your red pixel, there is no way to even know it is there unless that blue dot is essentially diffused enough to hit the blue pixel sites around it. This is why OLPF has to be done before the sensor because it is allowing additional material to be recorded that would be lost.
 
Well, Rick, even though, the chroma channels may be involved in some weird way leading to atypical behaviour (moiré and unbalanced highlight clipping). Or is the OPLF of bad quality, producing unwanted luma leakage / diffraction at diverse RGB wavelengths? These are of course speculations.
 
Is it just me, or is this whole conversation quite similar to the 'what native resolution are the HVX's chips' conversation back in 2007?
 
I like Barry's "The Wringer" test, but do not think comparing the HDMI output gives a clear picture (pun intended). The HDMI output of the AF100 is clean and can be recorded and played easily. The GH2 has some real issues with HDMI recording, which make the results highly dependent on how the recorded file is post processed. It is likely that the GH2 is suffering from some chroma issues related to HDMI recording and post processing. It would be better to do this comparison with the AVCHD recordings. I may be wrong, but I think most people will be using AVCHD recording and will forgo the complexity of HDMI unless they have very specific needs.

Second, I am going to stick up for the results from the German website. A few months back, people were discussing OLPF on one of the other threads. I responded that an OLPF is not the only way to eliminate moire and aliasing from a sensor signal. Arguably the best cameras in the world (medium format backs and the Leica M9) use NO OLPF over the sensor at all. They take their images with no additional blurring due to the OLPF. This leads to the superior resolution of these cameras. If there is any moire or aliasing present in the final photograph, it is left to the photographer to correct it in post processing using digital filters. This is a MUCH more flexible approach, but requires sophistication in post processing by the user. For people desiring the best possible results, this is the way to go. For the average user that doesn't do much post processing, it may lead to ugly artifacts.

Sophisticated digital filtration techniques can be a substitute for the OLPF. How does this relate to the AF100 and GH2? Everybody assumes that because the GH2 has a weaker OLPF than the AF100, it will have more moire and aliasing. That would certainly be true if there was no digital filtration. Just the act of binning the GH2 pixels into larger video pixels has a filtering effect, which will reduce moire and aliasing. This may not be all that is going on. Panasonic may have created a much more sophisticated algorithm for doing the filtering than simple binning. The GH2 does have marked improvement in moire and aliasing over the GH1, so something has changed other than sensor resolution. Here are the GH1, GH2, and AF100 test charts from the same German website for comparison:

http://www.slashcam.de/artikel/Test...udio---Bildqualitaet---Aus-dem-Messlabor.html
http://www.slashcam.de/artikel/Test...kompakte-Ueberflieger--Aus-dem-Messlabor.html
http://www.slashcam.de/artikel/Test/Panasonic-AG-AF101--24P---Aus-dem-Messlabor.html

Translated:

http://translate.google.com/translate?hl=en&sl=auto&tl=en&u=http%3A%2F%2Fwww.slashcam.de%2Fartikel%2FTest%2FPanasonic-GH1---Kit--Audio---Bildqualitaet---Aus-dem-Messlabor.html
http://translate.google.com/transla...kompakte-Ueberflieger--Aus-dem-Messlabor.html
http://translate.google.com/transla...asonic-AG-AF101--24P---Aus-dem-Messlabor.html

You can see the improvement of the GH2 in comparison. So, don't just judge moire and aliasing by the OLPF strength.

Personally, I would like to see Panasonic go the Leica route for its professional cameras. Forget about the OLPF entirely. I think Red either eliminates or uses a very weak OLPF on its cameras. Downrezing from 4K or 5K to HD is done in post with digital filters to take care of the moire and aliasing. Record all the sensor data and allow the user to process away unwanted moire and aliasing in post at an appropriate level. The rest of the time, we would get the much higher resolution that the sensor is capable of producing. That is why the weaker OLPF of the GH2 gives better resolution than the AF100. This also enables the ETC mode on the GH2 which the AF100 can't have with such a strong OLPF. Just my opinion, but others may like that they are guaranteed to have no aliasing or moire on the AF100 with a strong OLPF.
 
Last edited:
So Rick and I just compared the GH2 to AF100 with the test shots and some adhoc real world brick shots. I haven't had a chance to look at all of the footage yet. However, the one trend that I have seen so far is that it is all in the "details". As in the detail settings make all of the difference. We ran through the whole range from -7 detail to +7 detail with the AF100 and tested the GH2 from 0 to +4 sharpness. We also tested the intelligent resolution feature as well.

From what I have seen so far the AF100 definitely handles the test patterns better than anything else I have ever seen when you use -7 detail. Now it isn't very sharp like that but it definitely has less artifacts like that than the GH2 can achieve even with all sharpness settings turned off. As you raise the detail settings the AF100 begins to look a lot like the GH2. Personally, I think that the biggest difference is that the AF100 offers a much broader range of control(-7 to +7 Horizontal and Vertical vs. 0 to +4) than the GH2. You can achieve similar results with both cameras in real world situations. However, the AF100 will give you more range to cover those extreme situations that casual users properly never will encounter. Other than that I would be happy to use either one of them.

I will post the original files and also lossless joins of the files to Vimeo when I get the chance tonight. We tried to match the settings of both cameras as much as possible but there were slight differences. We were also pressed for time so we didn't have time to set the white balance properly on the AF100.

Does anyone know what settings for detail and sharpness the German test used? Without testing the whole range of settings for both cameras the test is pretty much useless.
 
I believe they used the default settings which are 0 for the detail and vdetail. I too looked quickly through all the footage we shot and, well, there is a lot to digest. Coming from a 7D background, I am not used to being able to see almost any of this stuff so I don't think I pushed the patterning as hard on the roofs/bricks to really test out the AF100. What I shot would have been HORRIBLE on the 7D.

My initial observation is that in our real world tests, both cameras did pretty well. That said, I felt the GH2 did show some more aliasing in what i saw, but also, it was sharper with false detail. But, this is just a quick look through. I need to dig a lot deeper and really get in closer with some of the shots to see what is going on.

I settled on -5 detail/vdetail with coring +3 for all our real world tests. As mpgxsvcd said, we were pressed for time so this isn't the best scientific approach we did.
 
I absolutely disagree that using no OLPF is the better solution. Without an OLPF, the sensor records detail that does not exist, period. As soon as you pass the nyquist limit on spatial frequency, you generate patterns, and removing those patterns is VERY difficult to do, especially in motion. The Red One, from what I have seen, uses a OLPF between the spatial frequency of green sampling and that of red/blue sampling. This is easily seen by taking a piece of screen material and recording through it at the right length. You eventually get blue and red moire in the screen, just as you would on a Canon (although VERY subtle, so not really the same). This seems to be the same OLPF strength a lot of companies take from what I have seen.

Yes, if you have the option to reduce the image from say 4k to 1080p then you've achieved the ultimate solution, that said, I'd still not want to record detail that didn't exist in the scene I was looking at. I would ALWAYS much rather add in an OLPF so that I reduce the spatial frequency to what I can actually capture. Red is preparing their cameras for 4K distribution so they aren't planning on down converting (much) in the future either with the Epics.

The Leica M9 doesn't have an OLPF because it doesn't fit, there is no room for one. Do a quick search of people using it and I've seen plenty of comments around the lines "Yes, it is very sharp, but it has a hard time with patterns" which of course, is why I moved over to Panasonic in the first place. I am tired of dealing with so much false detail like I was on my 7D. Have you looked at a 1:1 of Red Footage? It's not that sharp at 1:1, which when I look at 1:1 with my 7D (not sharpened) or the AF100(not sharpened), I see exactly like I would expect. Not so different.

There are a lot of questions here. The biggest being, we know that Panasonic likes to use pixel binning, but, they don't say how. Both the GH2 and AF100 are doing things different, which is easily seen when shooting the same scene. For instance, on the GH2, I see different aliasing between horizontal lines than vertical lines, whereas with the AF100, it seemed about the same for both. Removing the OLPF on the AF100 would make it look sharper, but the Moire patterns we see in the circles would DEFINITELY be worse, I have NO question on that. And worse, they'd be really hard to fix on moving images.

But even worse, look at it this way. If you take a camera without an OLPF and point it at a grid pattern of red dots. You can move the camera into a position where the red dots are NEVER over a red pixel. When you recorded, you'd never see any red. Now, if you place an OLPF over the sensor that spreads those red dots out, you'd be able to capture a tiny bit of red on the red pixel and at least show there is some sort of redness in the image. This is the point, no software in the world could tell you that the red was there if the sensor never saw it. While a theoretical example, this is why filters are going to fail. Trying to predict what MIGHT me there is a recipe for disaster.

Even further, ask anyone that has had to remove moire from a video shot. I can tell you first hand, it is NOT fun.
 
mpgxsvcd said:
...the one trend that I have seen so far is that it is all in the "details". As in the detail settings make all of the difference. We ran through the whole range from -7 detail to +7 detail with the AF100 and tested the GH2 from 0 to +4 sharpness. We also tested the intelligent resolution feature as well.
Click to expand...
Thanks, Rick and mpg, for your attention to the cameras' detail settings in running these tests. In the GH2's case, I'm particularly interested in how Film Mode Sharpness interacts with the Intelligent Resolution setting. As explained in the following article, virtually all digital cameras use some amount of sharpening to restore image details that are muted by the demosaicing algorithms used to produce RGB pixels from the RAW image sensor data:

http://www.imatest.com/docs/sharpening.html#Sharpening

Panasonic describes their Intelligent Resolution processing here:

http://panasonic.net/avc/lumix/compact/zs7_tz10/engine.html#intelligent_resolution

What I'd suspect is that the default midpoint Sharpness setting in both cameras selects something like what the Imatest article calls "standardized sharpening", and that the minimum setting is close to zero sharpening. If so, I'd expect the GH2's Intelligent Resolution feature to produce its cleanest results with Film Mode Sharpness setting at minimum. If it works as they claim, I.Res would selectively sharpen the edges of an image while leaving its smooth gradients unsharpened. I've tried this out on the G2 with I.Res at +1 and the difference it makes is significant.
 
Here is the first video. This is supposed to be a lossless edit but I am not sure how accurate they are with their claims on being lossless. I will upload the original individual files later. I just don't have time to do that right now though. Hopefully, this gives you some idea of what we saw.

http://vimeo.com/20049866

 
Last edited:
You must log in or register to reply here.
Back
Top