When small interaxials just don't cut it! - Page 31

Frank- Just watched the video on my Sony VW-90ES and it looked quite good. Motion was smoother as expected with the 240 Hz refresh rate. The convergence on the quadcopter was also good except where it came within a reach out and touch it distance. In my HT, I sit about 18 ft in front of the screen. The quadcopter began to diverge and show some minor ghosting at about 15 ft. in front of the screen. It was minor, however and had I not been looking for it, I doubt it would have distracted.

My viewing of it was off my computer SSD, over the lan, to my PS3. Then the Sony Projector was put in manual over under mode.

There are indeed two camps on this. You can decide if you believe the eye witnesses that participated in studies, or the engineer who conducted scientific testing.

It comes from the fact that you rendered the Top / Bottom video at "half" not full resolution. This squeezes the full res frame into half the vertical size of the full frame or 540 lines. There is a theory that since you are using two half vertical res images combined in the brain ( Brain Fusion Phenomena) that the combination is a sum total of 1080 lines in your brain.** The theory is nice if it held up to empirical data which it doesn't.* Sorry! Actual resolution testing clearly shows the lines of resolution are about half visual in the half frame rendering of a 3D image test chart. There was a demonstration of this at NAB. Applying the standard resolution test will show that the resolution is not full but indeed half even when the two images ( left and right) are combined in the brain per Joe Kane. One explanation as to why the full resolution is lost in a half frame rendering is that the lines of resolution are not actually squeezed together for later expansion by the monitor but rather half the numbers of lines are actually lost in the process of the half frame render. It gets worse! In an FPR screen the vertical resolution is again cut in half visual, resulting in a quarter resolution due to the rendering of half frame plus the FPR reduction of half the horizontal lines. The use of TB Half rendering for a 2K FPR passive monitor results in a 3D video that is 270 x 1920. This is why most 3D edtitors prefer to render SBS half for any final output that requires this 3D compatible mode. On a passive 2K screen the final visual resolution will end up being 540 x 960 which is a happier balance of both vertical and horizontal resolution. With an active screen ( LCD shutter glasses) the TB half rendering is 540 x 1920 and the SBS half ( common preferred render) is 1080 x 960. Again, the empirical data using standard 3D resolution test charts verifies these numbers. In addition, the math also would fall apart in the theory of brain merging of the left and right eye half res images, as the math would not increase the numbers of horizontal lines in a TB full.



* Joe Kane scientific testing of FPR passive resolution studies.
** Contradicting Joe Kane's scientific testing, DisplayMate Technology did their own study using real people who claim they see full 1080 lines using FPR panels and the company as well as LG believe this is due to ":Brain Fusion Phenomena"

These are the two theories, one is scientific and supported by empirical data while the other is based on eye witnesses. Eye witnesses are great for advertising and marketing by companies promoting FPR technology, but I prefer science, and regardless of whether the FPR is half resolution or imaginary full resolution, the picture looks great and I like the ease and comfort of the FPR passive as opposed to the higher resolution active shutter technology.

In my opinion, prejudiced toward the science, I feel the eye witnesses only imagined the image had higher resolution because the FPR panels are inherently brighter, higher contrast, and the glasses are more comfortable and cause less eye fatigue. In addition the FPR panels also offer less ghosting due to higher brightness and contrast. An eye witness may not be able to segregate the test subject of resolution as being human might just not be able to evaluate the resolution without being influenced by these other superior factors. In Joe Kane's studies, these other factors of superiority were carefully segregated so the resolution and math studies stood alone. So, when it comes to resolution, I side with Joe Kane but the synergism of the FPR passive advantages, wins my $.

In my procedure to create a rendering for 3D file import to PD10 I recommended using TB FULL with a custom frame size of 2160 x 1920. This would be nice to use on 4K FPR screen but will not fit on a 2K screen and the TV will squish the image, same for SBS Full or 1080 x 3840. I have done some experimenting with these full modes to display on the FPR screen by modifying the PAR appropriately. This method will work but unfortunately, other problems arise finding a monitor that will display non standard PAR's. I used Stereoscopic player for my experiments which allows one to fool the monitor with non standard PAR. So, my use of TB Full was only to move a Vegas rendering to PD10 for creating menus at the highest no loss of resolution. I could have used SBS full too but when I was working on this, Vegas 10 had rendering issues with SBS Full at 23.976 fps and 59.94 fps. These problems were absent with the TB Full renderings. I had no intention of ever displaying the TB Full rendering. It was used solely to preserve the full resolution of the original video.

If I understand you right, you are thinking about to combine two times 1080 60i to one time 1080 60p. The idea would make sense, if you are able to take the lines from the first field or L, and combine it with the lines of the first field of R. A fieldhas here 1080/2 = 540 lines, and so the idea is to render 2x520x1920 from the first fields ot the two streams to athe first frame of the 1080 50p stream - as top-bottom half.

The idea sounds nice. The only questionmark is how we are able to combines fields only. I do not know how that suits the internal workflow of our NLEs - I think they will deinterlace first and then calculate the new frame. Maybe that could be done with Avisynth in a simple way, but I am not familiar with Avisynth.

Maybe a way would be to render the 1080 60i footage to top-bottom-full still as interlaced format - and then deinterlace that to 1080 60p as top-bottom half. But I do not know if that works out really.

I am not sure if will be able by optical analysis to distriguish between a situation where a workflow deliveres interpolates and recalculates frames, and where not. The goal should be that the deincerlaced takes place in a way where we drop every every second line in the field - that is empty in 50i/60i anyway. But where we do not recalculate the other part of the field.

I am thinking about how to do that in Vegas - there could be different ways to end up with such a top-bottom footage. I am not sure if top-bottom-half with 1920x1080 as 50p/60p is the best way. It depends if the footage is recalculated or not before it is resized.

We could try to test if it makes a difference if we disable or enable the frame recalculation.

And in the project settings we have deinterlacing, what we typically use as blend or interpolate - I am not sure if here "none" would be the best choice or not.

We could also try to go for top-bottom full with 3140x1920, as 50p/60p or even as 50i/60i. But 3140x1980 can be done with the Mainconcept AVC encoder only.

So there are a lot of parameters that we can test to develop such a workflow. But the result could be to end up with a stream that maintains the full information in the footage, and that could be played back with a mediaplayer to a 3D HDTV. Interesting idea.

I have made a quick check with TD10 1080 50i footage.

Project properties:
1920x1080
uff
25,000 fps (for me it is a PAL project!)
3d mode: anaglyphic (can be anything else too, but must be a 3D mode)
8bit
good/best
deinterlacing: none

The project properties must not fit to the format we go for, but to the media we have in the timeline - to ensure a good preview capability. Mainlit the "deinterlacing: none" is relevant here, since this avoids that we deinterlace first and make then a resizing (that is still what I expect, I have no proof for that).

I have set the media properties to resampling disabled, but I think this is not important really.

More important - the render template and the encoder that is used:

1: Sony AVC/MVC encoder
I used the blu ray template 1080 50i, and mofidied it to following settings:

- AVC
- High definition 1920x1080
- high
- cabac
- 50,000 (double pal). User 60 (double NTSC here).
- field ordern none
- par 1,000
- bit rate 25.999.350

and in the project settings:
- render quality best
- top/bottom half


2. Mainconcept AVC encoder
- 1920x1080
- main
- 50,000 fps
- none (progressive scan)
- 1,000
- 1
- 50.000.000

and in the project settings:
- render quality best
- top/bottom half


The results seems to be quite similar - comparing the result with the original I do not see a difference really.

What did not work out was to render to top-bottom-full with the MC AVC encoder - here I see a shift in the luminance and a decrease in sharpness (only with top-bottom-full! For top-bottom half it was fine). Do not know why that happens, but I tend to use the Sony AVC/MVC encoder for that.

First of all I thought we were talking about RESOLUTION, not frame or field rate. Sorry if I didn't understand. Changing frame rate does not change resolution whether it is measured in lines or pixels.

The issue with RESOLUTION and FPR panels is a two-fold discussion when we are discussing video we create as opposed to commercial video simply viewed off a 3D BluRay disk. The first issue is the discussion of how FPR itself filters out half the vertical resolution. This is the debate I also referenced to Joe Kane vs. DisplayMate. The second issue is the rendering to half vs. full. Neither of these are affected by the number of frames per second or for that matter fields per second. BUT, you are correct in that if the odd and even lines could be stacked in the render contiguously ( is that the correct word here?) then it may indeed be as you assume, Frank, and be additive to the rendered resolution as a Brain Fusion Phenomena. Unfortunately the rendering does not stack the lines in this sequence. I agree with Wolfgang on needing some other render engine to achieve this different result.

In the DisplayMate argument, they make a claim that the FPR panel is a full 1080 lines because the brain combines the two images, left and right that are 540 for the full 1080. Yet they conveniently ignore the black lines in the FPR screen. So, if the full height is limited to 1080, and if the FPR screen makes every other line black, then the half the image is missing or screened out, so says Joe Kane. ( I agree) But DisplayMate argues that if you sit far enough away the black lines "disappear" and the remainder fuses together creating a full screen at 1080 lines. Joe says that is absurd, because sitting farther away from a screen does not double the resolution, if anything it makes the resolution even less, the farther away you watch.

Frank- If you are attempting to achieve better motion increments, then I stand corrected in where you are heading with this. In motion, the frame rate ( field rate) and the shutter speed and the monitor refresh rate all play a significant role. But, the pixels per frame or field do not. You don't need a cute quadcopter toy to test your motion theories. A simple fan with variable speed motor and a strobe-tach instrument should do the job. Resolution tests on the other hand can be measured with still frames. Joe Kane has developed a set of 3D resolution charts to isolate resolution measurements that are not confused by brightness, contrast, and other image qualitie including motion. DisplayMate did their testing using real world 3D BD productions claiming they had people view the letters on objects in the video and measured resolution by extrapolating the size of the letters in the movie to pixels per inch. When they called this science, I busted out laughing!

Maybe the question is, Frank, that you want to achieve better motion with controlling the rendering and how the stack of the lines work in the deinterlacing process ( again referencing what Wolfgang said) Here you may be onto something IF you can control the render engine and achieve something these codec engines may not have been designed to do.

You can put both the original footage in a track in Vegas, and put the rendered footage in the videotrack above - and then switch between both to see the differences. Works fine.

No, the footage was not likely to show quality differences.

The major drawback I see is another point: where do you play that footage? I have tested that with my Prodigy 3D, works fine here too, but the MVC-encoded files seesm to be better at my active plasma HD and that player. 1080 50p/60p does not fit to 3D Blu-ray. So the playback is the major issue. Maybe with the stereoscopic player.

Yeah, I get it. The use of some of the terminology before made me think you were trying to do something else. So far I haven't had an issue with 24p now with Vegas v11 but was worried with V10. I like 1080 60p and would shoot more of it if I could find a way to make it work in my display. Until the 4K LG TV's are out and affordable, I'll stick with 24p. I think they will only have the 4K in 55" and 84" this Fall. The refresh rate is also to be 480Hz ! Rumored pricing to be at ~$5000


In the meantime I have plenty of editing to do. My planned traveling is now over.

I look forward to seeing your Canon's in 1080 60p.

I watched the video on the Samsung C7000 LCD, since I returned the LG earlier today. I'm not quite sure what to make of it, except that it looks so lousy compared to what I was seeing with the LG. I also watched the copter video you posted yesterday, Frank. Despite the lack of jaggies on the Samsung, the dark image and severe ghosting made for some really lousy 3D viewing - double copters everywhere!!! I doubt I'll ever watch 3D on it again.

My Prodigy 3D plays that file - and it is also possible to switch to top-bottom in the player. Then my 3D HDTV accepts that as 3D. The other way works too - to replay that in 2D as top-bottom, and to switch to 3D on my 3D HDTV.
So yes, it can be played back.

Remains the other issue - quality seems to be good in the Vegas preview, but - at least on my active 3D HDTV with my 3D Prodigy - it is not as good as if I encode zu MVC. Do not know why that happens, and I have not found the time to test that excessive yet.

I guess I was looking for something new here. We visited all this over a year ago when getting the Vizio as an edit monitor was all the rage. Plus again when we were in debate as to how to render different outputs for transfer of files from Vegas to PD10 to author menus although 60P was not part of the process there because of incompatibility with Blu Ray standard.

And, If I may get back on topic, i.e. wide interaxial cameras-

I completed my evaluation of the two NEX 5 cameras in a twin camera SBS setup.

MY purpose for this test is to have access to different lenses to achieve shooting focal angles that is not easily obtained with my twin TD10's. I plan to shoot a stage where I cant get back far enough to get all in the shot with the widest zoom on the two TD10's. Normal (65mm) to about 150mm IA will work fine in this configuration. The initial testing will be with an e lens that will give me an effective focal length of 12mm on the NEX5n's.

The test I ran was with my Nephew's NEX5 and my NEX5n. This initial testing was to:
1. Confirm the ability to maintain sync with the two cameras because to my knowledge there is no way to add a sync device like the Lanc Shepherd to these cameras.
2. To verify the advantage of shooting 1080 60p over 1080 24p as the NEX5n does both these speeds.

To test #1 I planned to use the traditional clapboard sync method and compare both sound and image match between the two cameras both at the start of of the recording and near the end of a 10 minute long clip.
Results- First I aligned the left cam with the right cam's recordings quantized to frames with the timeline set to 60 fps using the beginning clapboard moment of close visual. The sound was about a third of a frame in delay. Next I compared the clap board visual again after the 10 minute recording. The visuals lined up exactly and the sound of the clap was again delayed by about 1/3 of the frame time length. The sync seems to be good enough for this project. I will not require any zooming and my wide angle lenses do not support zoom. But both lenses support auto focus and auto aperture. I plan to shoot shutter priority in this project.

Test 2 results- Had a surprise when we went to set up the nephew's NEX5. Being the older model the 5 only had one video mode, 1080 60i, so the option to shoot 60p and 24p had to be scrapped. I set my NEX5n to 60i and just gave a sigh.

Other surprises- When I compared the video of the NEX5 with the NEX5n, exposure and color were spot on but there was also a difference in the framing. The NEX5 did not appear to shoot as wide as the 5n so this presented a bit of a problem in pairing the two cameras in post. Using the pan crop tool I was able to zoom in on my 5n clip about 3-5 % to match up objects in the frame of each camera, then do an auto correct in the 3D stereo correcting with auto zoom and this fixed the focal length differences. I don't know if my nephew's 5 lens ( we both were using the same 18-55 zoom lenses for the test ) was accidentally pushed in a bit or what I found out later as the cause of this problem. What I discovered is that Sony changed the size of the imager in the 5N so that the 5 and the 5n are really different optics behind the lens. I didn't know this until after I got home. I'm no lens expert but I do believe that imager size difference can change the effective focal length of lenses.

This problem / surprise is good because I was planning to save some money and buy a used 5 for my project and now know that at least on two fronts this would be a big mistake.

I would use a camcorder setup like yours, Frank, with Canons but I don't believe they make a 12mm wide angle lens for that camcorder that does not exhibit vignetting and edge chromatic distortion. I think 24mm is about as wide as it gets and these are quite expensive. Plus, now I read where you advised me that they do not support 60p either. While I have not done any work yet with 60p, I plan to on my next Fireworks shoot with the TD10's.

Seems to be a similar experience with what I had when I tried to use my old TD10 and combine in with a newer one. Yes, to pair cameras they must be similar in a lot of points.

But I don't recall Sony changing the image CMOS mid camcorder model. They did between the 10 and 20 however. I'm still puzzled by your trouble with the two TD10's. The NEX5 to NEX5n was a huge change from what I now know. Its pretty well publicized. Some of the accessories don't even mount on the older 5, such as the OLED viewfinder.

I want to do this project but hadn't planned on what the new budget will be. NEX 5n's are selling for high used prices now. I guess a lot depends on how good sales are in my business between now and the end of the year.

That quadcopter is amazing! I liked the IA of 3.5". It kept the 3D effect natural.
I've prepared a video to show the effects of a mismatch between Left and Right timing on apparent motion. I captured at 60i with my Sony HDR-TD10, extracted Left and Right (using the MVC to AVI converter from 3dtv.at), and with VirtualDub converted to 60p (using odd and even fieds). I used VirtualDub again, with its motion interpolation filter, to arrive at 240p. The result was as if I had captured the moving orbs of the anniversary clock at 239.76fps!

I then used VirtualDub to harvest every tenth frame to get to 23.976fps (a frame "decimate" option in VirtualDub). But the point of harvest could be offset by 1, 2, ..., 10 frames, to simulate capture delays ranging from 4.17mS (1/10th of a frame at 23.976fps) to 41.7mS (one frame at 23.976fps). Even at 4.17mS, an effect on the motion of the orbs is apparent for my vision. Here is a link to the YouTube: http://www.youtube.com/watch?v=k_m4ETc-ydY

The video lasts just under 7 minutes. The smallest mismatch shown (4.17mS), begins at 3m 15 sec.


Unsycnchronized capture at 60i

I note that for those using two unsychronized cameras running at 59.94i for large interaxial 3D capture, the LEFT and RIGHT footage can readily be matched in video editing software to the nearest full frame quantization of 1/30th second, which -- assuming no drift -- would at worst be a mismatch in content timing of plus or minus 1/60th second. [Any mismatch greater than 1/60th second could be reduced by slipping the relative alignment of the LEFT and RIGHT clips by 1/30th second (or a multiple of 1/30th second).]

If attempting to match the clips by quantising to the nearest 1/60th second, the video editor will need to be able to cope with timing at the precision of individual fields. In subsequent rendering to avoid loss of motion detail it would in some cases be appropriate for the software to use the second field of one frame together with the first field of the following to create a new interlaced frame. If clip alignment is in fact possible in the editor to the nearest field, then -- assuming no drift -- the worst case mismatch of the video content would be plus or minus 1/120th second, or 8.34mS.

The perceptual effect (for many people) on the very regular motion of the anniversary clock is to alter the apparent trajectory of the orbs. The brain substitutes what seems a valid interpretation, drawing on a parallel from real life behaviour of moving objects when viewed stereoscopically in synch.

With irregular motion (such as the Blue Jays in Frank's earlier uploads!), a watery effect or a disturbing flicker can result, depending on the speed of the movement, the frame rate, the exposure time, and most importantly, the sensitivity of the viewer. In my video, the frame numbers in small white font may look "watery" in their last two digits because the Right and Left frame numbers differ by the timing mismatch. The brain is confused as the images presented to the Left and Right eyes differ in a way that has no parallel in real life.
Edited by MLXXX - 6/26/12 at 10:47pm

Frank, I am still not sure what this test shows me.

I mean, the important point would be to have each of the fields from 1080 60i footage shifted to the upper or lower part of the top-bottom frame. Without recompression if that is possible - and for sure without new calculation/interpolation of the fields.

So, if you take the 60 positions in a second in the timeline, and if you render that to 60p, you will end up with the situation that every second frame must be interpolated. Simply, because you have 30 fields for the top, and 30 fields for the bottom part of the top-bottom-footage. So maybe it would be better to render to 30p, and disable resampling. And set deinterlacing to none.

We can only hope that this works. I am not sure if your test can show that this has worked out really?

I have to agree with Wolfgang's comments. I am unclear on what advantage has been identified.

Perhaps the reason the overall effect looks good using half top and bottom, rather than half side by side, is simply that horizontal resolution is fully preserved. For display on a passive LCD screen that can only show half of the vertical lines per eye this may be a particularly good fit: the passive set has full horizontal resolution capability and half top and bottom mode fully exploits this capability.

If using bob deinterlacing for half top and bottom then motion is maintained at 60fps, but vertical detail is slightly compromised because the detail from the odd lines is taken from slightly different parts of the image than the detail from even lines. However in practice with sophisticated deinterlacing algorithms and image sharpening, the loss in visible resolution may not be all that noticeable. I observe this sometimes when I watch a standard definition digital TV broadcast of what was a high definition feed to the studio (e.g. of tennis). The image can be surprisingly sharp [though there is often some aliasing, e.g. of the tennis net, or lines on the court.]
Sorry about that. In the editing I tried to insert text in places and at apparent distances that would blend in with the content of the video. But yes it is a fairly intense video to watch all the way through, with its multiple images of the clock and its reflection in the mirror.

I would be interested in any reports confirming (or not) the optical illusion of the orbs appearing to spin outwards and then inwards. For my partner and me, the effect is quite pronounced. Perhaps for Don and his wife the effect would be less noticeable. I can see the effect even with the minimum timing mismatch of 1/240th sec or 4.17mS, which can be observed beginning 3m 15 sec into the video. I conclude that unsynchronised stereoscopic capture at 60i would not be well suited to shooting objects that move in repeated patterns, fairly rapidly; at least not for people sensitive to 3D timing mismatches.
Edited by MLXXX - 6/27/12 at 8:54pm

The last time my wife had any interest in these tests was watching Frank's birds and squirrels eating peanuts. She was so inspired by it she went out and bought a bag of peanuts, and a bucket just like Frank's, sat it in the middle of our yard and watches the squirels with my spotter scope. Sorry, Jeff, but watching that clock spin away doesn't seem to hold her interest quite the same.

For my use, I will be shooting motion in my ultra wide angle twin NEX 5n's using a 6" IA and the 12mm lenses. I got my first test lens in today and will be laying out the geometry to see if this is wide enough to capture my wide stage. If not wide enough, Sony has a fisheye lens I can use but that one has considerable geometry distortion and I wanted to avoid that. The geometry distortion of the 12mm is not really that bad. The motion involved will be the fountains at the Bellagio so sharpness is not a real concern. My other application for 60p will use my TD10's and here I have the Lanc Shepherd to sync the TD10's. wide angle is not the most important requirement. In fact I will probably be shooting at 5x telephoto.

Purely technical videos would not interest a lot of people, though the optical illusion might be of interest to some.

Don, the Lanc Shepherd, does it actually force the TD10's into synch, or merely report on the instantaneous difference in synch in say milliseconds?

Thought this was the most appropriate place to report that I'll have the stereo base extender for the Panasonic Z10000 in a few weeks. It's supposed to be about 240mm, significantly wider than similar devices by Cyclopital3D, but also supporting a wider angle setting for the camera. I'm anxious to see what that combination can do.