Current State Of Crosstalk/Ghosting - Page 3

What firmware release do you have installed on your C8000 I understand that there was a new one today or yesterday?

I've updated to the 1022.1 firmware that was released on June 28. I tried the picture correction and set the 3D viewpoint to 0. I still see the same amount of crosstalk in the scenes

I see absolutely no evidence that any crosstalk is baked into the MvA encode.

If anyone wants to prove this to themselves, all they need to do is get hold of a Panasonic 3D Blu-ray player, force it into side-by-side output mode, and watch in 2D on their display. The separate left and right decoded images can then be easily seen (albeit horizontally squeezed).

I have done this and inspected several scenes which people have reported as problematic (the bridge scene, the chapel tower, the alien gun, etc.) I see absolutely no crosstalk in the separate left/right image pairs.

James,
How can you tell from the SbS images that there is no crosstalk?

I had the 40 inch Samsung LCD (7000 series, I believe) for about one week and watched several of the world cup games. The ghosting was so bad on many distant shots that each player looked like a group of three moving on the field.

Strangely enough the ghosting would become less obvious as the the TV had been in 3D mode for 20 minutes or so. I started running the set for 20 minutes before I would even try to watch it.

Anyway, after a week, I returned it and bought one of the Panasonic plasma 3D TVs and have had no significant ghosting issues at all.

Before the firmware upgrades, the ghosting on the Samsung sets was so bad, they were, IMO, virtually unwatchable. The upgrades have all but eliminated that problem on the plasma sets, and I'm assuming the LCD sets, too. It was a sync problem.

Are you saying that when the Panasonic 3D Blu-ray player is forced into side by side mode, it takes the two eye views of a 3D Blu-ray disc, compresses them to 960x1080 and outputs them as a single 1920x1080 side by side stereo pair?

Absolutely. The left/right views are stored as separate streams on the disc. The player needs to decode them and composite them into whatever the final output format is. In the case of the Panasonic deck, this is either frame packed, checkerboard, or side-by-side.

My point was that by viewing the side-by-side output on a 2D display, one can easily inspect the discrete left/right views.

That is incorrect. If that was the case then a 3D BD would have a file size that is 100% bigger than a normal BD. But it doesn't. It's file size is approx 50% larger. The reason for this is the AVC-MVC codec. There is redundency in 3D images so it is not necessary to store both L & R views. The missing 50% is created by the decoding.

http://research.nokia.com/files/3D_Video.pdf

Is this a potential area for ghosting/crosstalk?

Simply by virtue of the fact that I observe no crosstalk baked into the images when viewing them SbS on a 2D display. I'm not trying to be snarky--I just don't know how else to answer your question.

Yes, I'm aware of that. Technically the 3D data is stored as interleaved base and enhancement Transport Stream (TS) data. 2D players "see" only the base data, while 3D players can interpret the enhancement portion as well.

I've seen quotes ranging from 30% to 60% for the overhead of the delta, but the 50% figure has somehow become ubiquitous whenever the subject is discussed. I guess the point is that only a portion of the extra data needs to be stored.


I suppose anything is possible. But any decoding solution that introduced crosstalk would be a poorly engineered solution indeed. I believe my side-by-side tests on MvA prove that it is at least possible to decode 3D BD without introducing crosstalk artifacts.

So is it possible that crosstalk is created by the interpolation of the missing data by the MVC decoder?

In side-by-side, if viewed as a 2D frame, you see the left image squeezed on the left side, and the right image on the right. Side by side, as the name says.

This allows to look at the left and the right image separately. If the ghosting would be encoded, then it would show up in those images. If not, then they are clean.

I was editing my original post as you were writing...

But as I said in the edited version, I'll concede that it's possible (but not likely IMO).

I've seen side by side formatted left/right views extracted from 3-D blu-ray releases, and can confirm they are "ghost free".

D'oh! Of course. I knew the Panasonic players could output in different formats, but I haven't used any format except native 3D with my Samsung plasma. I knew they would output checkerboard, and that's one way my sister would be able to watch 3D Blu-ray discs on her new rear pro DLP.

That's the great thing about 3D Central - people coming up with these simple tests that should be obvious to everyone, but for some reason escape us until pointed out by someone like yourself, James. (And I own a Panasonic 350 player. )

I'm curious if the Samsung 3D Blu-ray players can output any other format except frame-packed Full HD 3D. I know they don't do checkerboard output, so you can't hook them up directly to the DLP rear pros. You need the Mits adapter.

What you all are saying is that if either the right or left SbS frame for a scene that has cross talk is viewed separatly that if the side created from the +delta MVC data has crosstalk it will show up when displaying the SbS frame, correct?

I have been under the impression that it the SbS frame created from the MVC +delta content contained too much of the data from the 2D data in the MVC content that the problem would only show up when displayed as 3D.

I don't mean to speak for James, but what he is saying is that the Panasonic Blu-ray player pulls both streams off the 3D Blu-ray disc (both the left eye view and the derived right eye view, both full resolution 1920x1080), then tosses away half the vertical resolution so that it can combine the two resulting 960x1080 images into a single 1920x1080 frame. That frame, which appears in 2D as two squeezed images, side by side, can be examined for any traces of crosstalk. If you see none, then it means the no crosstalk exists in the encoded 3D image.

I never believed that would be the case anyway, but James' technique demonstrates clearly that MVC encoding does not compromise the right eye view with crosstalk. The crosstalk culprit, as we all thought, is either the glasses or the display, or both.

Yes.


Once the left/right image pairs have been decoded, their pedigree (that is, the format in which they were previously encoded) can have no further effect on them. If the left and right images are 'clean' upon decoding, and crosstalk is witnessed when viewing them in 3D mode, then the culprit for the crosstalk cannot be the encode.

What I don't understand is how you can see if the individual side frames have cross talk since they are not being displayed with another frame to have cross talk from. Which is why I would always expect them to look clean when viewed separatly.

I will pick up sony 3d player tonight; can't wait to start some experiments.

Both the right and left images from the 3D Blu-ray are visible, as a single side by side, vertically squeezed pair (viewed as a 2D image - no 3D engaged on the display itself). Even with half the vertical resolution gone, it would still be obvious if any crosstalk occurred as a result of the MVC encoding process.

Viewed as 2D, it's conclusive evidence that MVC is not a crosstalk culprit. View such an image in 3D (by engaging side by side mode on the TV) and you will see what sort of ghosting the display and/or the 3D glasses introduce.

I am not understanding how you have decided that your test is conclusive proof.

The only way to get crosstalk is when in 3D mode. You have to display BOTH images on the display in some 3D format. It is impossible to get crosstalk in 2D mode.

THIS isn't 3D. This is a 2D frame using the SbS 3D configuration.



This is 3D.

My understanding is that someone proposed that crosstalk might be encoded into the derived right image itself, and that it would be visible no matter what. That's what is being discussed here. If a ghost image of the left frame were a part of the right frame, the best way to see it would be to look at ONLY the right frame. Looking at both frames in 3D would never isolate the right image, and therefore would prove nothing.

As far as I'm concerned, the idea was an unlikely proposition from the get go, and it's case closed. James' technique shows that it has no validity. Also, as he says, if the encoding were guilty of doing that, it wouldn't be much of a codec to use for 3D in the first place.

The question was whether or not there was ghosting in the encoded video, as opposed to the displayed image. Looking at the left and right frames separately, as the side-by-side image allows, can be used as proof that there is no ghosting in the encoded MVC stream.

So, if the 3D is ghost-free up to that point, then the ghosting has to come from somewhere else down the line: The TV's processing, the display panel, the glasses, etc.

Are all video decoders the same? That function is done in the SoC. Are Broadcom, Sigma Designs and Panasonic's Unipher SoCs equal in performance?

I think most of us are in 100% agreement with your assessment:

The SbS 3-D views clearly show this.

I have re-evaluated my contention that a current generation MVC encoder can cause ghosting by means other then by have the ghost image actually part of a frame and determined the it can not. Since I also don't think a current generation MVC encoder can create individual frames containng ghosts I now don't think any MVC encoded content contains ghosts.