(Page 1 of 2 pages for this article  1 2 >)

Filed under: *VIDEO* •3D •Audio •Business •Cameras •GentryMedia Sister Sites •HDSLR •ProVideo Coalition •Hardware •Post Production •Stereoscopic •

HPA Tech Retreat 2011 Day 4

On this fourth and final day of the 2011 Tech Retreat, we learned about standards activities, 3D ghosts, camera arrays, automated audio “recognition”, a new method for making film protection masters, how bending a cable affects its performance, and a whirlwind tour through TV Tech history. Also: the death of tape… for real this time?

These Tech Retreat posts are barely-edited stream-of-consciousness note-taking; there’s no other way to grab all this info in a timely manner, get it published, and still get enough sleep for the next day’s sessions!

I often use “distro” as shorthand for “distribution”, and “b’cast” for “broadcast”. You have been warned.

What’s Happening at a Standards Organization Near You - Peter Symes, SMPTE

Brief overview: lots going on; Wendy promoted to Exec VP; Hans Hoffmann of the EBU now Engineering VP.

Annie Chang and Howard Lukk chairing the IMF working group (see yesterday’s coverage of IMF).

SMPTE working on: ACES, ADX, APD; 3D Home Master (single-point delivery master, like IMF for 3D); 3D disparity map representation. BXF: Broadcast Exchange Format (traffic info to automation systems), XML based, over 80 companies participated, being widely adopted; BXF 2.0 under way. ST 2022-x, Video over IP, in conjunction with VSF.

Synchronization and time labeling: Successors to color black and timecode. Why base digital timing on color subcarrier frequencies? Slower progress than expected. IEEE 1588 developing in right direction. More IT-based than previously expected. May wind up with IEEE 1588 with a bit of video-oriented metadata added. May work out that all gear—60Hz, 50 Hz, 24fps—happily synchronizes to the same signal. Should be low-cost, too.

Time Labeling: a new timecode? Won’t replace SMPTE 12M TC as it’s far too widely used, even outside of the industry. Originally designed as what can be recorded on the edge of a quad tape, and what can be displayed on nixie tubes, neither of which is very important nowadays! Nice to have more info; to maintain original camera time along with program time. Need good input from the post community; HPA and AMPAS organized a day-long meeting for discussion and feedback; very productive. Q: Do we need a frame count, or just a high-precision time reference? A: We NEED frame counts! More info “when we have a strawman.” If we’re replacing something 30 years old, we’d better get it right.

Other projects:

Reference displays: need to get EOTF (electro-optical transfer function) right. CRT was incredibly good; other displays emulated CRTs. Differences between pixel-matrix displays are trickier; harder to get consistent display rendering.

Lip sync: perpetual challenge. And when we went to ATSC, which has timestamps to ensure A/V sync, errors went from a few frames to a few seconds!

Dolby E for 50Hz and 60Hz; Archive Exchange Format; 25 Gbit/sec fiber interfaces; accessibility-times text standards for D-cinema and broadband; ongoing D-Cinema, MXF, metadata work.

Three new tech committees: broadband media and television; media packaging and interchange (IMF and 3D Home Master); media systems, control, and services.

The standards ecosystem:

Digital Leader: SMPTE has a set of TIFFs and WAVs to make leaders for DCDMs (head and tail leaders).

DPROVE: set of 48 DCPs for for testing / setting up theaters and screening rooms: 2D and 3D, 2k and 4k, multiple aspect ratios, etc.

Professional Development Academies: monthly webinars, free to members, low-cost to non-members; check ‘em out on SMPTE website.

Upcoming programs: Digital Cinema Summit at NAB; regional events; conference on stereoscopic in NYC in June; SMPTE Australia in July; annual tech conference in Hollywood in October. Next year: global summit on emerging media tech, with EBU in Geneva in May.

Redesigned website, new member database, digital library with all SMPTE pubs since 1916 online.

Lots happening, please participate!

Q: what about the centennial in 2016? Working on big plans.

Measurement of the Ghosting Performance of Stereo 3D Systems for Digital Cinema and 3DTV - Wolfgang Ruppel, RheinMain University of Applied Sciences

Wolfgang is a “long-term survivor of D-cinema projects”.

Ghosting is the perception of leakage (or crosstalk) of one channel’s info into the other channel’s eye. Due to imperfect separation of channels, and a main reason for 3D discomfort.

One way of getting perfect separation.

This sort of display is actually used today for 3D X-ray viewing!

Ghosting can be characterized as:

R’ = R + k x L

L’ = L + k x R

where:

R’ = what the right eye sees

L’ = what the left eye sees

R = right eye content

L = left eye content

k = leakage amount

That would be simple, simply subtract k x the wrong channel. But this won’t work for subtracting a bright ghost from black since you can’t go negative. Also, k not constant; depends on angle, brightness, color, ambient light, etc. 

How to measure L and R? Linear light, gamma corrected code values? We’re using code values; perception based on lightness, not luminance. Measured crosstalk may be 10% in linear light levels, but is 44% in code values (a proxy for perceived lightness).

Test chart:

Left Eye chart. Right Eye chart basically swaps white circles for black, and vice versa.

There’s both white ghosting and black ghosting; there’s also color-shifting in ghosting.

Ghosting suppression level: looking for single-value suppression levels; used charts to see what level of bleed was.

Charts for looking at suppression levels.

Looking at color changes due to ghosting, too:

Color shifts in black parts of the image due to ghosting.

The white ghosting color shifts were much lower.

Spectroradiometer measurements of ghosting correlate well with perceived ghosting.


Ghost suppression differs a lot depending on color, with different systems having better luck suppressing red crosstalk, others blue; the suppression levels differ depending on ambient light, too. Brighter 3D shows more ghosting than dimmer 3D.

Discussion: please don’t use code values, because they differ between systems, use L* values instead. A: Doesn’t really matter since we can’t control brightness levels across different venues, so code values vs. L* doesn’t make any practical difference.

Photorealistic 3D Models via Camera-Array Capture - John Naylor, TimeSlice & Callum Rex Reid, Digicave

Using purely passive state of the art camera arrays. (These are the folks doing the multiple Canon rigs to shoot Mirage surfing spots; see also how they did it).

3 points of merit: determinism (must get the shot), getting every camera to trigger at exactly the same time; resolution; number and layout of cameras.

Stabilization doesn’t matter.

Array costs ($400,000?) can be amortized over non-3D jobs. Customer benefit: buy in day rate, no R&D.

Determinism: multiple cams of the same type have generally the same latency, but there will be variance:

Resolution: the more pixels the better:

Demo: a 360 degree timeslice shoot in Paris. Two hours to rig the cameras using a pre-built support system.

Image interpolation with software from The Foundry.

Test with Digicave, Phantom slo-mo plus 40-camera timeslice.

Digicave: free viewpoint media production. Focus on 3D interactive content. Tech enabled, but not tech focused. 3D body scanning.

Samples of the 40-camera test with Timeslice; 3D models derived from the pix:

Full wireframe model created from multiple images.

Color panted back on; model can be viewed from any angle interactively.

3D modeling data-extraction software developed over 3 years by in-house PhD guru.

Demos of interactive 3D models, full fly-around from arbitrary angles.

Want to go to motion work; RED cameras because 12 Mpixel images are the minimum required.

Qs: how many cameras used for the 360 shoot? 60. For the all-round view of Callum? 36 cameras. Some touch-up needed on the top.  Interpolation is good, but better not to have to do it (two hour step). Higher-res cameras more important than having more cameras. Tested with motion? 16 HD cameras, but they were only 2 MP, and the results look like it. Difficult to ensure consistent color / exposure across cameras.

New Audio Technologies for Automating Digital Pre-Distribution Processes - Drew Lanham, Nexidia

Dialog search, Dialog-to-transcript alignment; audio conform (synching multiple audio tracks).

Originally out of Georgia Tech in 1997, 10 patents, in Avid’s ScriptSync, PhraseFind; also an FCP plugin.

Dialog is a rich source of metadata. 100x faster than realtime, language identification. Able to search for any word of phrase once the index is created. Can use boolean operators and time values.

Demo of Nexidia SearchGrid: 130k hours of YouTube clips, searched in about two seconds for the word “china”. Downloaded videos; scanned and indexed, stored metadata, dialog, ID of the video for later recall.

Note the amount of hits returned by audio search compared to metedata search.

Bar graph under video is timeline of search term occurrences in the video.

Dialog to transcript alignment: Avid ScriptSync. Lets you navigate through program by text, or finding text corresponding to a part of the program. Also find places where text and dialog differ in content. Identify changes between versions. Create rough cuts by selecting text. Generate timing lists for ADR, captioning.

Full-length movie took 2.5 minutes to process on a MacBook Pro.

Audio conform: audio based similarity analysis. Can detect “are these assets related?”, can detect drift, gaps, incorrect content, distortion or dropouts. Believe it’s more consistent and accurate than human review. Output is XML based so easy to use elsewhere.

All apps are software based; highly scalable; SDKs on linux, Mac, Windows.

Discussion: search is tweakable for recognition threshold, to trade off false positives and total returned hits. Anyone using it for captioning? Not yet; people are looking into it. Certainly helps for automating the timing process. Phoneme-recognition based, or using dictionaries? Phoneme based, not dictionary based, so it works on slang, which isn’t in the dictionary. Code is about 50 MB, language packs are 10-20 MB.

Archiving Color Images to Single-Strip 35mm B&W Film -

Sean McKee & Victor Panov, Point.360 Digital Film Labs

[Presenter had different name, but I was in the middle of a discussion and didn’t catch it. -AJW]

Visionary Archive. Two methods:

RGB+: HD or 2K full RGB to 3 quadrants of a frame; 4th used used for metadata recording. Added pixel-accurate alignment patterns around each quad to ensure protection from shrinkage and distortion (10-pixel grayscale sinewaves). Also black, white, midgray reference to preserve level (and color) values.

RGB Intra-mosaic: super a Bayer pattern, separate out two Green quads, plus 1 Red and 1 Blue.

Tested on Arrilasser2, Aaton K, Cekloco, etc.; Kodak 2234, 2238, 5269, Fiji RDS 4791.

“Rosetta stone” frames at the head so that someone with no knowledge of the format can easily decode it:

Why film? 300 year lifespan, and it’s immediately comprehensible: just hold it up to the light.

Workflow: run files through the Visionary encode software, data-record out to B&W film, store the film. Safe, simple, should last a long time.

Tests: less grain than YCM masters, no re-registration required, warped film not a problem (alignment pattern for de-warping), creative intent preserved (white/gray/black patches preseve color and tone values).

Free source code for un-archiving. Hoping to make it a standard component of film scanner software.

Audio: digital coding in an analog wrapper. Oversampled 2x, all manner of redundancy. 16 channels per frame in a quadrant.

What’s next? 3D stereo archive element; metadata integration.

Playback at the Tech Retreat compared original, versions created from YCM master and from Visionary Archive master. VA master version showed better color fidelity than YCM masters, even after color correction. Said to show lower grain, higher res, but I was in the back of the room so I couldn’t see such finely detailed information.

Discussion: with the 4-quad mosaic method, where’s the audio? Use a separate filmstrip. Mosaic gives higher visual quality than 3-quad version. If the audio is digital, doesn’t that break the discoverability of it? The problem was how do you put something analog on film and have it survive scratches; audio is very sensitive to distortion. Thought of an analog guide track. In any case, human-readable audio is a lot harder than human-readable pix. Nit-pick: most film-out systems already include grayscales and calibration wedges. Yes, but not per-frame. How does the small quad allow you to claim lower grain? The 3-quad system doesn’t, but the mosaic system does; since you’re not shrinking/enlarging the mosaiced samples, there’s no grain enlargement. Any sense in going larger than quads, like 3-perf sequential? Yes, looked at it, but lose the economic advantage: area vs. money!

Next: Bend radius; Schubin and history; final thoughts…

(Page 1 of 2 pages for this article  1 2 >)

Thanks a lot for the coverage. It’s amazing to see so many topics covered, and to know that some people are tackling the problems we’re going to have to deal with in years to come.

The move to file only delivery is going to be hell on earth…

Posted by Corbor  on  02/21  at  03:56 AM

---

It’s already hell on earth for a bunch of these folks!

One of the things that strikes me, as I sit back and reflect, is that last year at HPA we saw some “best practices” presentations. This year, while there were roundtables with that aim, none of the presos were about “best practices” but, seemingly, just about surviving the flood.

The great terror—or the great comfort, depending on your viewpoint—is that nobody knows what “best practices” actually are; everyone is feeling their way into this brave new world (“oh, Brave New World, that has such complications in it!”) just as blindly and as hesitantly as everyone else. When folks like Disney and WB and Starz are making it up as they go along, you either feel comforted that you’re not the only one going though this trauma, or you’re mortified that if these folks haven’t got it all sussed, how the heck are we (and anyone else) supposed to figure it out???

Me, I’m a wee bit comforted that my confusion and consternation are shared by the Big Guys. But I’m weird that way.

Posted by Adam Wilt  on  02/21  at  10:36 PM

---

It’s like everybody has its own custom made cookbook and even though those folks at WB and Disney seem a bit confused i’m not worrying about them. I mean they can afford the LTO-5 recorders.
The thing is at the lower end of the spectrum where i usually work, this is going to be total chaos. And trying to convince people of the best course of action is going to be increasingly difficult when i’m not even sure what the best course of action is, for finding that sweet spot between cost, quality and stability.

Posted by Corbor  on  02/22  at  07:05 AM

---

Name:

Email:

Location:

URL:

Smileys

Remember my personal information

Notify me of follow-up comments?

Submit the word you see below: