AVCHD Torture Tests

BobDiaz

The Professor
With all the talk about what AVCHD can or can not do, I though it might be interesting to try and break the CODEC. The toughest thing I could think of was to shoot the waves coming in and out at the rocky beach of Malaga Cove in Palos Verdes. I used a Panasonic HMC40 to shoot these shots with the shutter set at 1/250 of a second and the iris said F5.0 (it's really at F2.8 with an ND filter).

I shot from the top of the cliff down to the beach below (see red dots on photo). I estimate the distance to be around 120 to 150 feet. The camera was set to infinity focus and the zoom at maximum. Part of me wonders if I shouldn't have double checked the focus, because it might have been 94 or 93, rather than 95. Still the footage seems reasonably sharp.

Cliff.png


I've placed the files on VIMEO for anyone to download (assuming you have a VIMEO account). These are the native files from the camera, so look them over closely to see if the CODEC falls apart or not.

http://vimeo.com/bobdiaz/videos


Now I know that a Panasonic AG-HMC40 is NOT an AF-100, but this is Panasonic's AVCHD CODEC and is likely to be very close to what the AF-100 would be. I can understand if Panasonic improves on the CODEC, but I can't see them making it any worse.


Take a good look at the files and post your comments here...



Bob Diaz




 
The AF-100's sensor should be able to resolve more detail then the HMC-40's sensor though, so its not going to be taxing the codec as much. Barry did a good comparison with an external recorder hooked up to an EX-1 I believe (hint someone should dig up the link), that would be a better comparison since we are all expecting the AF-100 to be a very sharp camera.
 
In answer to several questions...

The HMC40 is sharper than the 150, 170, and the 200. I'll call on Barry to back me up or call me wrong about that. :) I selected a fast shutter speed to record as much detail without the motion blur we would expect from the moving water. While I'm sure when Barry runs the test charts out, the AF-100 will beat the HMC40, but the upper limit is something under 1080 lines and the HMC40 came in close 800 or 900 lines give or take. Sorry, but I don't remember exactly; so there's not a whole lot more detail to feed the CODEC.

The files are the native AVCHD H.264 files that come off the camera. Not every system can play them.

I want to stress that the analysis is very subjective and what might be OK to one will be unacceptable to another. This is why I posted the source files for anyone to download and look at.

The correct way to look at the image is at normal playback. A still image fails to capture the dynamics of the image and can make the compression noise look worse. Also, the iamge changes frame by frame, so I have to capture at a point that looked typical for noise.

Starting at 6 Mbps, I can see the compression noise even at 1x size:

HE_6_Mbps_1x.png


By 2x size the image really falls apart:

HE_6_Mbps_2x.png


In order to see the problems as bad as the 2x size, you would have to look at the TV with a magnifying glass. In reality, you should look at the 1x size from about 3 to 4 feet from the monitor to get the same effect as viewing from a regular TV.

As I said, the still image makes the compression noise look worse, but even a still at 1080/24p looks reasonable from 3 to 4 feet away:

1080_24P_1x.png


A closer look at 2x reveals that the image held up reasonably well:

1080_24p_2x.png



I would expect that if the CODEC would fall apart, the moving water at the beach should break it. To some degree, there's some compression noise, but it doesn't seem to jump out.




Bob Diaz
 
It seems hard to deduct codec performance when the camera is losing so much true detail due to chromatic aberration
 
I definitely think that Barry's torture test in XDCAM EX vs AVCHD is a better example, a range of fairly common types of setups, and mainly focused on high-motion, which is really where a Long-GOP codec will struggle. In those tests AVCHD pretty much pwned XDCAM EX, and EX is a very popular and widely accepted format (and will involved in the new Sony camera in some way?)

The CA really does 'muddy the waters' in this test to some extent. But it's interesting to look at. Even at the lowest bitrate the AVCHD is performing better than I'd have expected.
 
Thanks for posting Barry's test, with so much on the Forum, I missed that one.

An A/B test is always better because you can compare one thing to another. In my case, I only have one thing, the HMC40.

However, without question, the images from the EX at 35 Mbps (MPEG 2) vs AVCHD at around 21 Mbps show that AVCHD is better than 35 Mbps MPEG-2.


Bob Diaz
 
Back
Top