When small interaxials just don't cut it! - Page 7

Excellent test, Frank. Obviously, the Canons blow away both the JVC and Sony in terms of 3D depth. No surprise there - that should be all the proof anyone needs. Both the single camcorders dramatically flatten space, while the dual Canons maintain the separation front to back. The difference is night and day. It makes me want to get on eBay and find another Canon to match the one I have (HF10). That's real 3D.

I also noticed that all three 3D systems exhibited stuttery motion when objects move across the screen swiftly. You can see it as you move screen right to left, and as the kids run a few feet from the camera. I'm actually relieved to see this. I had begun to think it was a problem with the JVC alone. It's not. I wonder if this is just an artifact of the conversion from 60i to 24p. This stuttering doesn't appear when I watch the raw footage straight from my TD1 on my 3D displays.

I really enjoyed watching this exercise. Thanks again for doing it. Do you have more tests planned?

Look up multi cam function in Vegas. Not sure if it works with 3D but worth a quick study.

I'm on the road near Columbia, SC

Regarding 60i and the stuttering motion for people walking by: I observed this as well when doing 60i to 24p with Vegas. My solution was to use After Effects CS5 to perform 60i to 24p (round trip using Sony YUV, Matrox I-Frame HD, or Cineform- no conversion loss will be visible). Some folks report that the yadif plugin for Vegas also works well.

To fix the JVC footage: apply a small blur and/or blur followed by an unsharp mask (if available in your NLE). Film looks better than digital as there is no aliasing (which creates an artificial sharpening effect). Blurring will help remove aliasing, and unsharp masking can bring back up apparent sharpness by removing low frequency information (without adding back in aliasing).
Additionally, a slight increase in contrast and color saturation will help the JVC more closely match the Canon and Sony footage.

For low light noise- Neat Video is your friend.

At Petri's suggestion, I've used Yadif to remove most of the objectionable stuttering from my JVC problem clips. Ones that I had considered unusable (like what we see of people close to the camera and moving laterally in Frank's videos) become quite acceptable. Good news. What still baffles me about the MVCtoAVI process, though, is that I don't see the stuttering (or at least not nearly as much of it) in the JVC mp4 (MVC) clips when I play them in Windows Media Player (which presumably is playing just the left eye view of the MVC pair). As soon as they're split, the stuttering appears (or is much worse) in that same left eye view. Any idea why that might be, John?

This part of the test has me puzzled. The JVC image looks less saturated and less contrasty than the Sony. In previous clips that I've seen, the opposite seems to be the norm. I've never found the JVC wanting for color or contrast. Frank, you also say that you think the JVC looks overly sharp, but in this test that doesn't seem to be the case. I'll look forward to seeing the other footage. This test has been very interesting. I'm still hoping my friend gets the Sony this fall. I'd love to see how the Sony would perform at Shaw's Garden.

I probably won't do it, but I've been looking on eBay for another Canon HF10. I'll probably just wait until next year's models arrive to make a change, but if I can pick up one cheap enough, it would be fun to play with a dual camera rig.

Make sure the output is interlaced and the frame rate is preserved (29.97 or 59.94- usually 29.97 is correct if interlace is maintained). Try with Cineform and make sure interlace is preserved (one side effect I need to report to Cineform is the last line is green (bug)). Otherwise, the quality is excellent with Cineform (I'm using Low to save disk space; a 'Difference' test showed Low as good enough for my footage (can reprocess and rerender later if necessary (treat initial files like proxies)).

I tested Matrox I-Frame HD at 102Mbits- looked OK (interlaced, 29.97).

(Realized after I wrote this that it's more appropriate for an editing thread. Oh, well, it's easy enough to skip it. )

I've maintained interlace throughout the process. This is a quirk I can work around now, so it's no longer a top priority to figure out, but one of these days I'd like to know exactly what's going on.

Could you point me to the software you use to measure differences in file integrity? I have test clips that I can use to eyeball differences (fine details around tree leaves, motion clips), but the software you're describing is obviously more precise.

I've tried the Matrox I-Frame HD codec at different bitrates (again, thanks to Petri's suggestion) and it looks good at 100 mbps, but at 120 mbps I can see virtually no difference between it and Cineform at "High." And it's about 20% smaller. Since it saves time in post (if I don't have to do FirstLight corrections), I'm going to use it for my next project.

You've mentioned a couple of times about doing titles in Vegas after editing in PPro. Is that because of rendering time in Premiere if you do the titles there? I prefer Premiere's titler to Vegas (maybe it's a matter of my greater familiarity with Premiere, but I think it's more feature rich). If rendering time is not an issue for you, why not do the titles in Premiere, too?

If almost sounds like the old HDMI signal level bug that used to haunt me at every turn, but I've never experienced that sort of thing in Vegas. What I put in is what I get out in terms of contrast. It's just the JVC clips that are washing out, and not the others?

All my Vegas Garden clips are Cineform intermediate files (HD-Encoder 2, "High" setting). I've never experienced that problem. Any other system changes lately? Sometimes the cause of a problem seems completely unrelated. Once, when someone stepped on a butterfly in China, my video started playing double time in Premiere.

Very interesting approach to comparing the 3 systems. As for the washout problem, here's a thought - try a different codec in MVCtoAVI and see if the problem persists. (Split a file and render something short in Vegas as a quick test.) Could be that the registry cleaner whacked some file that was needed by Vegas for the codec you used when you split the JVC clips.

I was fishing for an "other side of the tracks" joke, but they were all lame.

Sounds like a Vegas bug to me.

You can do the 'Difference' test with just about any NLE: Drop the original clip on track 1, and drop the transcoded clip on track 2, above. On the track 2 clip, under Opacity (PPro, for example), select "Difference". You can now scrub through the file and see any changes. Sometimes you can't see the changes at this point- that's good. To see more, select both tracks and right click and select 'Nest' (PPro). Add the 'Gamma Correction' Video Effect to the nested clip. Increase brightness to see the differences more clearly.

The difference test can get more technical using statistics- PSNR (related to mean squared error): http://en.wikipedia.org/wiki/Peak_signal-to-noise_ratio

To compare visual differences, go back to the un-nested clips and rapidly toggle visibility of the top track. This test is good enough for most people- the difference test with gamma cranked up is helpful in understanding what kind of losses are occurring and peace of mind.


You can do titles in PPro (and AE). While PPro & AE titling capabilities are much stronger, for quick titles in 3D Vegas is sufficient. After contacting Cineform tech support for some Neo issues (all fixed after re-registering all the plugins), I found out the 3D Parallax Video Effect is for titles. Create the 2D title, drop it into the sequence, add the 3D Parallax Video Effect- and it is rendered in 3D, with the ability to adjust parallax. After the help from Cineform tech support, Neo is looking much better

Thanks, John. For my Shaw's Garden video, I did all my titles in Premiere. I have a couple that still need work. To finish one, I'm going to keyframe the parallax effect, so that the title can recede into the distance more realistically. The Premiere numbering convention for parallax control bears little resemblance to the one used in Vegas.

Frank,
Off subject questions can you provide a thread link the contains the following info.
Which make/model of 3D TVs can display multiple 3D windows and what 3D format(s) vsn you use with them?
When using multiple 3D windows is there any problem with syncing the glases to all of the windwos concurenty or can you only sync to one window at a time?

Walford

Hey Frank, that's always painful. 0x80010105 looks like a COM error (probably related to codecs (in Microsoft COM format DLLs) on your system. If Vegas processes video first, then audio, it could be related to audio codecs. Could also be a memory leak, out of memory/disk space error, or a pointer related bug (trashed/garbage pointer to a COM object that when called results in this type of error when called between processes (vs. an access violation when in-process (0x80000005)).

You might try the PPro CS5.5 demo with MVCtoAVI and Neo (together, less than the cost of a 3D camera; all have demo versions) and see if that workflow is more stable on your system (tradeoff is it takes up a lot more disk space and the extra work prepping the files (not too bad as everything can be batch processed). Once done, editing performance is much faster (faster than real-time) and on my system PPro renders at 6-10x faster than Vegas). On supported CUDA hardware, PPro is overall much faster than Vegas (if interactive and rendering performance is important).

You could also try just MVCtoAVI and Neo with Vegas: will edit much faster and might be more stable (the Vegas MVC code is new- could be a cause of stability issues).

Per Cineform tech support, a future tool from Cineform (Neo/HDLink, etc.) will support MVC directly (will be able to skip MVCtoAVI step).

I've been missing Premiere so much the last few weeks. Recently, Petri posted that the Matrox codecs were available for free download (to be used as an alternative to Cineform). Last night, I went back to Premiere to check out the workflow with Matrox MPEG2 I-Frame HD clips.

I used MVCtoAVI to convert the JVC mp4 files at 120mbps (indistinguishable to my eye from Cineform High, but using only 80% of the bitrate). I kept the audio in the right track for compatibility with Vegas down the line. I dragged the clips into the Premiere timeline, with left on the top and right on the bottom. I then grouped the left/right clips (Ctrl-G). That way, when I trim I keep things in perfect sync. Since Premiere doesn't have to deal with the second track of video for realtime playback, it's just like editing any other HD project - that is, it's smooth as butter. I was even doing the test on my C: drive.

With this workflow, I doubt I'll have the realtime rendering performance John is getting with his 12 core and Quadro 5000, but I plan to upgrade my CPU (presently a Core i7 920) to boost rendering speeds. I'm going to wait to see what sort of performance I get from my GeForce 460 before I think about a Quadro video card. I've done the CUDA hack with the GeForce cards in both my systems. HD editing is a breeze, with several HD streams and multiple titles playing in realtime on my existing Core i5 system (only 4 cores, no hyperthreading). I'll still have to go through Vegas to render to 3D Blu-ray, but I'm going to use this workflow for my next project and see how it goes.

It felt so good last night to see my the timeline play without any hiccups.

You should get real-time performance on your GTX 460. It's only slightly slower than the Quadro 5000 (something like 115Gflops vs 113Gflops). From earlier testing for PPro editing, a GTX 285 is about as fast as the Quadro 5000 for interactive as well as final GPU assisted rendering. The GTX480 and up are faster.

For final CPU-based rendering, yeah- cores, RAM, & clock speed...

Gotta see pictures of that rig!!