Puzzled by AVCHD vs MJPEG (Screen Shots)

[Exact]

I'm shooting 720/50p and have set the following:

AVCHD 38, 40, 42 as per the stable thread
MJPEG 384, 330, 300, 274 / 24, 24, 24, 24 - again as per the stable thread

Every one seems to be raving about the MJPG settings, but it doesn't matter what I do, AVCHD seems far superior to the MJPEG, at all ISO settings. So I'm wondering if I could have done anything obviously wrong

There is also some apparent banding on both @ ISO 800 and above - is this normal?

Here are some example stills (100%):

Also, for those who have not played with either the GH1 or Neat Video, this is the same AVCHD frame with Neat Video (noise reduction) applied.

 

[sammysammy]

I guess, the difference is small, and in my eyes its small, i liked the mjpeg more, but it does not mean the avchd was not good, its very hard for me to decide, so they are both close..just being picky..

 

[blazer003]

I think that the MJPEG has a little more color information, and if they can get 1080p out of the MJPEG (last I've heard they have not), then it might be better.

Personally, I agree with you. In almost all situations (at least right now) it seems to me that I like the image results from AVCHD much better.

 

[wturber]

Your point of focus shifted between the two samples. Your detail comes down to black and white text. You are shooting ISO 800 and the resulting noise increase will certainly mask some of the differences especially in banding. Your shot isn't of the kind that will show up one of the MJPEG's biggest advantages - resistance to posterization/banding.

If you want to see that key difference, find a situation where AVCHD gives you posterization/banding, and then shoot a comparison MJPEG.

 

[Exact]

I guess, the difference is small, and in my eyes its small, i liked the mjpeg more, but it does not mean the avchd was not good, its very hard for me to decide, so they are both close..just being picky..

The problem to me is that the MJPEG version is :

a) noisy (more noisy than the AVCHD version)
b) the colors are wrong
c) exposure is not only different, but just plain wrong too.

The AVCHD version more accurately paints the scene, both in real life and what was seen on the LCD prior to starting to shoot. The MJPEG version is nasty compared to the AVCHD version. To my eyes, on my monitor, there is a huge difference between them.

Settings didn't change on the camera, except switching between AVCHD and MJPEG. Everything else remained the same.

Your point of focus shifted between the two samples. Your detail comes down to black and white text. You are shooting ISO 800 and the resulting noise increase will certainly mask some of the differences especially in banding. Your shot isn't of the kind that will show up one of the MJPEG's biggest advantages - resistance to posterization/banding.

If you want to see that key difference, find a situation where AVCHD gives you posterization/banding, and then shoot a comparison MJPEG.

The bad news is that the point of focus did not change. I checked and rechecked these results 4 times to make sure that the focus didn't move. The MJPEG just screws up the test on the battery for some reason. The camera is locked down on a very sturdy tripod. I also placed a wide rubber band around the focus ring to make sure it didn't move.

One of the reasons I set up this test in a more controlled environment was because I had previously been testing outside with people, cars and trees. The MJEPG was a lot more noisy and had banding, whereas the AVCHD seemed fine. So I naturally started questioning if I'd done the hack correctly.

In case it makes any difference, I'm using a Panasonic 8GB Class 10 card.

 

[PappasArts]

I'm shooting 720/50p and have set the following:

AVCHD 38, 40, 42 as per the stable thread
MJPEG 384, 330, 300, 274 / 24, 24, 24, 24 - again as per the stable thread

Every one seems to be raving about the MJPG settings, but it doesn't matter what I do, AVCHD seems far superior to the MJPEG, at all ISO settings. So I'm wondering if I could have done anything obviously wrong

No reason to be puzzled.

Your MJPEG example doesn't have the same focus, and is suffering from distortion from wide open aperture, and the AVCHD shots are not. . Mjpeg is not processed in the way that the AVCHD signal path is. The MJPEG shows you the noise, while AVCHD has some NR like circuit going on in it. MJPEG should not go above 800iso, and performs beautiful at much lower ISO's. Remember that 800iso on the GH1 is actually above 1100iso.

Here are better examples of MJPEG performance for you to gauge with..

AVCHD vs MJPEG - 24PN vs 24P over 60I etc
http://www.dvxuser.com/V6/showthread.php?t=214735

1080P MJPEG VS 1080P AVCHD Stills
http://www.dvxuser.com/V6/showthread.php?t=213935

MJPEG high bit rate footage-stills & comparisons.
http://www.dvxuser.com/V6/showthread.php?t=212609

Lowlight MJPEG 1080P frames
http://www.dvxuser.com/V6/showpost.php?p=2017000&postcount=93

1600iso with a Nikon 35mm @ F1.4 @ 1/40th in 1080P high bitrate mode
http://www.dvxuser.com/V6/showpost.php?p=2019902&postcount=109

Test blow up of 720P Hibit rate MJPEG MODE to 1920 and beyond.
http://www.dvxuser.com/V6/showpost.php?p=2009705&postcount=29

Anamorphic frames/lowlight- MJPEG
http://www.dvxuser.com/V6/showthread.php?t=214489

MJPEG high bit rate footage-stills & comparisons
http://www.dvxuser.com/V6/showpost.php?p=2011043&postcount=44

MJPEG 70mbits+ 1080p settings
http://www.dvxuser.com/V6/showthread.php?t=213364


Pappas
http://www.pbase.com/Arrfilms

 

[Exact]

No reason to be puzzled.

Your MJPEG example doesn't have the same focus, and is suffering from distortion from wide open aperture, and the AVCHD shots are not. .

I know you say that .... but if the focus had changed between AVCHD and MJPEG then it must have some how magically reverted to perfect focus when I ran the second AVCHD test, then magically gone out of focus in the same way when trying MJPEG, and back and forth 4 times, all with an elastic band around the focus ring on the lens to prevent it moving

The focus point, aperture, lighting, shutter speeds etc were all the same. The only thing that changes was the codec.

Now, having said all this, I am willing to accept that MJPEG was not the appropriate codec to use in this instance, and this is the kind of info I am looking for. When to use AVCHD and when to use MJPEG for best results. However, I was not expecting such a big difference on this one setting.

 

[PappasArts]

elastic band around the focus ring on the lens to prevent it moving

The focus point, aperture, lighting, shutter speeds etc were all the same. The only thing that changes was the codec.

Now, having said all this, I am willing to accept that MJPEG was not the appropriate codec to use in this instance,

Close up shots would have never had that degree of difference.

You had an error in some way with your test.

This is a focus/optical/settings error of some kind, not a codec error.

Pappas

 

[Exact]

I will re-test again tomorrow.....

 

[wturber]

And note that the MJPEG image shows more detail in the lens labeling than the AVCHD image. If it was a CODEC issue, then you'd expect the AVCHD image to be sharper in both locations.

I don't set focus using anything to lock the focus ring. I simply use the 10x focus mode. It is easy to get exact focus that way. I recheck between each shot.

Also, why test for detail with a fast lens wide open. At least stop the lens down to f/2.8 to get some sharpness. Better yet, go to f/5.6 if you want high detail to test. Fast 50mm primes are notoriously soft wide open.

BTW, you gave the settings for the MJPEG but not what the data rate was. What was the MJPEG bitrate?

OH, and on a final note, there isn't a lot of detail in this scene and it is a locked down scene. Shallow DoF and a very soft lens. Not much for the AVCHD CODEC to trip on. Both of these things favor the AVCHD CODEC in addition to this scene not having areas that really stress the posterization/banding issue.

And to be clear, I'm not on an MJPEG bandwagon looking to "defend" MJPEG. In fact, I'd love it if AVCHD would consistently outperform MJPEG. But it doesn't seem to. It just does in some cases/scenarios.

 

[Exact]

And note that the MJPEG image shows more detail in the lens labeling than the AVCHD image. If it was a CODEC issue, then you'd expect the AVCHD image to be sharper in both locations.

Hmmm. I will recheck this on more shots.

I don't set focus using anything to lock the focus ring. I simply use the 10x focus mode. It is easy to get exact focus that way. I recheck between each shot.

I used 10x focus at the start, in the middle and at the end. There was no difference in focus, so no changes we made at all.

Also, why test for detail with a fast lens wide open. At least stop the lens down to f/2.8 to get some sharpness. Better yet, go to f/5.6 if you want high detail to test. Fast 50mm primes are notoriously soft wide open.

That's fine. It was a partial test of the lens wide open as well. I want to see how the bokeh was being rendered 'and compressed', so this was a deliberately shallow DOF.

BTW, you gave the settings for the MJPEG but not what the data rate was. What was the MJPEG bitrate?

Exactly which tool is every one using to measure this bitrate? I just want to be sure we are all using the same tool so that when I find the bitrate you can be sure we are talking apples and apples.

OH, and on a final note, there isn't a lot of detail in this scene and it is a locked down scene. Shallow DoF and a very soft lens. Not much for the AVCHD CODEC to trip on. Both of these things favor the AVCHD CODEC in addition to this scene not having areas that really stress the posterization/banding issue.

You make a very valid point about not tripping the AVCHD codec, but that is part of what I am trying to figure out - when to use each codec. Since this was a static scene I expected both to be high quality. As it is, the MJPEG version is nasty - with lots of noise. This was 'unexpected', since people are talking about 'stills' quality.

And to be clear, I'm not on an MJPEG bandwagon looking to "defend" MJPEG. In fact, I'd love it if AVCHD would consistently outperform MJPEG. But it doesn't seem to. It just does in some cases/scenarios.

I will indeed be doing many more tests. Thank you for your response. It's good to get other people's perspectives on things, especially when they are different to your own. It's only be questioning things that we increase our overall knowledge.

I will be back with more shots !!!

 

[Chris Light]

i do my own tests, almost daily, and all's well. are screen shots really showing us anything? we're shooting motion video, right? sorry if i come off as rude towards folks posting stills from video....but stills comparisons are not the same as video.....love to see these analyses on moving pictures so i can compare them with what i've got. reading what i just wrote, i sound like a d!ck. not my intention.

 

[timbook2]

I am trying to do MPEG vs AVCHD comparison tests but have one question :

the files created in MPEG HD are NOT 16.9 but come out as 4:3!
Mediainfo and QT both say 4:3.....and QT of course plays them distorted ...

any tips? the manual clearly states that MPEG HD is 16:9 but even playback inside the cam is squeezed 4:3!

I only set the changes in FW for MPEG like this:

MPEG size>1920
width 1440
height 1080
color sampling 4:2:2

this is PAL so I didnt touch any other settings...

AVCHD with the b settings works beautiful!!! I only have class 6 cards so for now I stay with that.

 

[timbook2]

I just reset all MPEG settings, upped new FW and now its 16:9 alright, so which settings did the 4:3 change????

 

[Exact]

i do my own tests, almost daily, and all's well. are screen shots really showing us anything? we're shooting motion video, right?

Yes, but the video looked nasty - which is why I started looking closely at the stills.

sorry if i come off as rude towards folks posting stills from video....but stills comparisons are not the same as video.....love to see these analyses on moving pictures so i can compare them with what i've got. reading what i just wrote, i sound like a d!ck. not my intention.

Say what you think, I've no problem with that. We're here to share ideas and experiences and as long as things stay on the technical side and don't get personal then it should all be good stuff.

 

[Nitsuj]

i do my own tests, almost daily, and all's well. are screen shots really showing us anything? we're shooting motion video, right? sorry if i come off as rude towards folks posting stills from video....but stills comparisons are not the same as video.....love to see these analyses on moving pictures so i can compare them with what i've got. reading what i just wrote, i sound like a d!ck. not my intention.

Well sure however you can't really show the quality in a streaming video as much as you can a still. After compression the video doesn't look near the same as it does on your system. So stills are a viable source to compare.

 

[Filmkid2003]

Hey Nitsuj, I think we have the same SDHC card. Patriot 16GB Class 10. I noticed you said you haven't had any problems with the hack. Can you please tell me what your settings are? I loaded the "D" settings with Native 24p checked and I get error messages every now and then saying "Cannon record motion picture due to write speed limitation" or something like that. Also, I can't play ANY of my AVCHD clips when I press the playback button. I get an error message that says "Cannot play this format"! Can you please tell me what firmware hack settings you are using? Thanks.

 

[wturber]

i do my own tests, almost daily, and all's well. are screen shots really showing us anything? we're shooting motion video, right? sorry if i come off as rude towards folks posting stills from video....but stills comparisons are not the same as video.....love to see these analyses on moving pictures so i can compare them with what i've got. reading what i just wrote, i sound like a d!ck. not my intention.

It is a fair point, but looking a the footage after its been crunched again by Youtube or Vimeo has its problems too. So it comes down to posting original files and downloading.

 

[Nitsuj]

Hey Nitsuj, I think we have the same SDHC card. Patriot 16GB Class 10. I noticed you said you haven't had any problems with the hack. Can you please tell me what your settings are? I loaded the "D" settings with Native 24p checked and I get error messages every now and then saying "Cannon record motion picture due to write speed limitation" or something like that. Also, I can't play ANY of my AVCHD clips when I press the playback button. I get an error message that says "Cannot play this format"! Can you please tell me what firmware hack settings you are using? Thanks.

I have a 16GB Class 6 Patriot actually. I am using the latest ptools patch. My settings are set to B as C was giving me too many write errors. I also have 24pN checked. I actually had one write error today but that was the only one in 2 days of shooting a lot of video.

 

[wturber]

I used 10x focus at the start, in the middle and at the end. There was no difference in focus, so no changes we made at all.

Maybe not intentionally, but something shifted.

I overlayed the MJPEG still with the AVCHD still. I then carefully aligned them (set the top image to "difference" in Photoshop and shifted until I had the blackest scene). I also made a curves adjustment to the MJPEG so that it had the same look as the AVCHD (pulled the blacks down a bit). I set the image back to "normal" and then clicked the top layer on an off.

Doing this makes the focus shift obvious. You can see the focus shifting all over the image. You can see the back half of the battery becoming sharper and you can see how the far background is less blurred on the MJPEG image. You can also see the shift in focus on the woodgrain surface. Somehow the focus shifted. The magnified out-of-focus area demonstrates this. The Photoshop technique I just described demonstrates this. Repeat this Photoshop comparison and I'm sure you'll end up agreeing that somehow the focus shifted.


There are lots of ways to measure bitrate. Most players and editors will report bitrate and you can always just calculate it based on the file size and clip length. You can also download a little app called "bitrate viewer."