Home of Professional Cinematography since 1996
Published : 5th July 2004
Noel Sterrett wrote :
>Unfortunately, I have found that user bits are hardly ever set correctly. >They are also not well integrated into editing equipment.
User bits, that is 'date' and 'roll number' are the only way to keep same TC numbers apart. You'll always have TC numbers showing up more than once.
What are you suggesting to do?
Sound mixer, etc
Karl Lohninger wrote:
>User bits, that is 'date' and 'roll number' are the only way to keep same >TC numbers apart. You'll always have TC numbers showing up more >than once.
>What are you suggesting to do?
I don't know what Noel is proposing, but what's done now is to simply name the directory that the files are in, much as you name a roll number on tape. The names can be based on whatever you want, but the logical way to do it is to name the master directory for the "roll number" and/or shoot date, and the "subdirectories" for scene, and then take.
For final conforming, a combination of the timecode based EDL and the "negative cut list" (which contains roll number, scene, and take) is used to uniquely identify all frames. There is no actual standard for this that I know of'; all current DI vendors have their own ways of combining this information for the final assembly. All file based systems have a hierarchical directory structure - they have to.
To me, the complicated thing to do is to stuff all this metadata into each and every file during live recording (i.e., production) and make it all work in a way that conforms to some sort of standard and is reliable. That, to me, seems quite a ways off, both in terms of reliability and simply deciding on a standard, regardless of what the AAF people seem to say.
IATSE Local 600
> I don't know what Noel is proposing...
You and me both...
For years I have been struggling with "timecode" which is overly complicated and at the same time inadequate for the task at hand, much less what is coming down the road.
Ultimately, post is a database management problem, and the first rule of database management is to have a unique ID for each record. It seems to me that if an individual sample (in this case image) is given a unique name, and the conditions under which it was recorded (time, place, device, etc.) then that sample can later be synchronized in time and space with other samples (cameras, audio recorders, motion sensors) without ambiguity. While it may not always be possible to record all the information one would like, time (including date) and device (camera A) are an excellent start. Like a Kodak Keycode, time + device provides a unique number for each frame recorded. Unlike Keycode, useful information (time) is also embedded in the numbering scheme.
Although Scene, Take, etc. should be recorded as well, they do not present a unique numbering solution as does time. It would seem more appropriate to store that information within the file itself or as a summary of a sequence of time/device stamped frames. I personally prefer using XML for this since it is widely accepted and can actually be read as text, rather than binary computer speak.
Perhaps SMPTE could define a Samplecode which we could all implement and get back to worrying about character and content rather than dropping our frames every 1000 or so.
Mike Most writes :
>all current DI vendors have their own ways of combining this information >for the final assembly. All file based systems have a hierarchical >directory structure - they have to.
[Lucas] Not necessarily. Look at the way the Quantel iQ or the Apple iPod organize media. You have the option of folders, but much more powerful is the concept of a saveable, multi-column metadata sort. In many projects, I've come to a point where a folder structure that worked at the beginning of a project quickly becomes unwieldy and/or irrelevant.
What are folders anyway? They are ways of organizing subsets of media. If every piece of media has several metadata identifiers, I think it's much more powerful to have scriptable, sortable searches that can be saved. Those saved searches can be called whatever you want -- folders, bins, libraries, etc. Look at the Smart Playlists on an iPod. Dynamic searches that are constantly live, and filter new material into locations based on a set of rules. Once I started using those, when I now get locked into folders, I feel like I'm taking a step back.
[Mike Most] To me, the complicated thing to do is to stuff all this metadata into each and every file during live recording (i.e., production) and make it all work in a way that conforms to some sort of standard and is reliable.
[Lucas] There are a few basic accepted standards of keeping track of timeline based media in post -- Keykode, Timecode, and File Numbers. There are subsets like Edgecode, Source and Record Timecode, etc., but every project comes back to these three basic categories of metadata. It's a question of reliably marrying those three metadata sets throughout the course of a project. For that, we have EDL, FLX, ATN, FTL, AAF, AFE, ALE, CTL, and a few others that I'm sure I'm forgetting. Each one of these has its place and function. Manufacturers have the Keykode/Timecode relationships pretty established. For me, the task is how do we incorporate file names/numbers into the mix?
[Mike Most] That, to me, seems quite a ways off, both in terms of reliability and simply deciding on a standard, regardless of what the AAF people seem to say.
[Lucas] Call me an eternal optimist, but I don't think it's that far off. Give me these applications: EDL Manager, Avid Log Exchange, FilmScribe, EDLMAX24, EDL2AAF, AAF2EDL, AAF2XML, and Excel, and I can go from just about anywhere to anywhere else in post. Granted, that's a lot of widgets, but it's there. The stuff we deal with is complicated. I don't think it will ever be bullet proof, but the methods are there now. It's a question of simplification.
This is of course strictly for timeline translation. Start talking about migrating color settings, LUTs, Color Cubes, DVEs, and things get significantly weirder.
[Noel Sterrett] Ultimately, post is a database management problem, and the first rule of database management is to have a unique ID for each record.
[Lucas] Metadata is fluid. What is true for a piece of media one day is not true the next. Maybe the timecode that was digitized is wrong. Maybe the 16mm Keykode in FlexFile should have interpreted as 40-count instead of 20-count. A unique identifier on one system is not the same on another system.
There is obviously a beginning point -- the origination camera and the format it is recording to. But from there, I think the key is being able to fluidly navigate between list sets and metadata sets. That ability makes translating between different systems easier, and it makes troubleshooting much easier when something goes wrong.
(disclaimer: I work for Quantel)
Michael Most wrote :
>what's done now is to simply name the directory that the files are in, >much as you name a roll number on tape.
Disclaimer first - "I work for nucoda - we make visualisation and conforming tools for the digital film market".
The concept that Michael describes above is used for what we would call file name conform - where the name of the folder that the file is in can be used to identify the "reel" name from an EDL.
However we also do what we call "Header conform" where the software will look for the information specified in the EDL - in the DPX header (i.e. timecode and reel name) The DPX header can contain a 32 character name, which can be used to uniquely identify every scene/take. I was actually demonstrating this to someone today, whilst recording from a Viper.
So with every shot uniquely named, ie tk1, tk2, tk3, etc the timecode is quite up to the task of identifying the shots - it also means that all the frames can be mixed together in one big folder if you want, because it is the meta-data in the frames that we read to do the conform.
>and make it all work in a way that conforms to some sort of >standard and is reliable. That, to me, seems quite a ways off
Works today - and the meta-data is the SMPTE DPX standard.
nucoda - Product Specialist
Nigel Hadley wrote:
>So with every shot uniquely named, ie tk1, tk2, tk3, etc the timecode is >quite up to the task of identifying the shots - it also means that all the >frames can be mixed together in one big folder if you want, because it is >the meta-data in the frames that we read to do the conform.
Well, in a 2000 foot AB reel, that would mean you would have 32,000 files - and that's without handles. Seems like an awfully unwieldy approach, and it also seems like it would take the software some time to read 32,000 headers and build whatever database it's going to use to index them. In any case, putting all the files in one directory prior to a conform seems like a dubious working method, even though the software itself can handle it. Seems to me that separate named directories is a lot easier to keep track of and a lot easier to deal with, particularly if those files are ultimately going to be copied to a "conformed" directory for color timing, either on the Nucoda system or elsewhere.
BTW, Nigel, if you could send me your contact information, I'd appreciate it. I might need to talk to you about something in the near future.
IATSE Local 600
>So with every shot uniquely named, ie tk1, tk2, tk3, etc the timecode is >quite up to the task of identifying the shots
The problem with using scene, take, etc. as an identifier is that they must be manually input and are, therefore, subject to error. Not that they should not be included in the file header or even the file name, but just that they are less than ideal as a unique file ID. Device name/number + timestamp can be unique and automated in an way that is much less prone to input error. The use of time in this manner has proven reliable in a variety of computer systems.
The problem with using multiple directories in a hierarchical file system is that it is often difficult to maintain the hierarchy. If the hierarchy breaks down, which I have often seen happen, the file names themselves are not unique and conflicts occur.
Kodak automatically generates unique Keycodes for film. Why settle for less with digital images?
Noel Sterrett wrote :
>Kodak automatically generates unique Keycodes for film. Why settle for >less with digital images?
True, but film is always catalogued and tracked by a combination of key number, camera roll, and scene and take numbers, even after it's broken down. The fact is that because film is a physical entity, it is not possible to be "sloppy" about storage and tracking systems and still find what you need. In many ways, it's much easier to keep track of a physical asset than an electronic one. However, I understand your point.
Years ago (around 1987 or so) at Lorimar we decided to try cutting and assembling negative from video EDL's (foolish at the time, but we did it anyway). We investigated using Aaton code as the basis for this, figuring that each individual frame was uniquely identified by using a combination of time code, shoot date, and camera ID.
I still think that's all you really need, provided the date and camera ID are properly entered and the time code generator is working properly.
IATSE Local 600
Michael Most wrote :
>True, but film is always catalogued and tracked by a combination of key >number, camera roll, and scene and take numbers, even after it's >broken down.
Yes, but the same can be done with electronic images, at least where that information is embedded in the frames themselves. It would be rather simple (in software), given even a huge single directory to create a database cataloguing and sorting the files by any fields (scene, take, etc.) included in the filename or file. DPX files provide a wealth of fields to store metadata, including a "roll-your-own" header.
The main point is that if we insure a truly unique record ID for frames (filename) all sorts of cataloguing and database management systems are possible. If we don't, it's going to get very messy.
>by using a combination of time code, shoot date, and camera ID. I still >think that's all you really need, provided the date and camera ID are >properly entered and the time code generator is working properly.
Exactly! This scheme works perfectly. There is one little weakness that, I believe, doesn't play a big role in practice and can be overcome if the software behaves intelligently: If you have a take that wraps over midnight it might get sorted into the wrong "shoot date" or software might have problems with the time code wraparound. (All assuming time of day timecode).
I see your point in wanting to generate a unique id and a timestamp. I agree that this would be desirable.- However, like I tried to point out in a previous post: For a timestamp to work you need to synchronize all CineRAMs on set in a perfect manner - plus you need to synchronize them to other equipment: To tape recorders, audio equipment, timecode slates.
There already is such a synchronization mechanism today. It is limited, complicated, counter-intuitive etc. But it works and is the standard : Timecode.
We might have used CineRAMs on a current project, but it involves 4 simultaneous HD streams and can't be done without an external timecode input.
The timestamp is a very good feature, because it allows you to move the frame easily to the correct "shoot date" folder and generally adds a searchable tag that can be used for a production database. It also gives you a good cross check when you shoot time of day timecode.
Still a timecode input is absolutely essential for the syncing of several CineRAMs and other equipment.
Lin Sebastian Kayser - CEO -
"For years I have been struggling with "timecode" which is overly complicated and at the same time inadequate for the task at hand, much less what is coming down the road."
With all of the recent discussion, I will quote below, an email I solicited from Evertz on the possibility of creating a Time Code with Date Stamp. If any of you has questions about the process, I am not the expert, but you can go to evertz.com and contact them directly for further information :
"Our 5010-GPS and 5600MSC+GPS units can put the date into the LTC user bits according to SMPTE 309M format. Our HD9010TM HD Timecode Master unit can then take in the LTC and insert it into RP188 ANC timecode in the HD bitstream for you to decode. Unfortunately the HD9010TM does not lock to GPS directly...but with either the 5600MSC+GPS or the 5010-GPS it can be accomplished .
If you are operating in a 29.97 world, then there will be an approximately 2 frame error per day. This error will get corrected by a hard rejam to the GPS time once per day at a user specified time. So you should always be within 2 to 4 frames worst case.
That's life in the 29.97 world! If your running at 25.00 or 30.00 FPS then there is not drift and everything stays exactly locked all the time (except for leap seconds - where there is a jump of 1 second)
GEORGE C. PALMER
HD and Digital Imaging
>"If you are operating in a 29.97 world, then there will be an >approximately 2 frame error per day."
That's true because the drop frame correction is not enough and studio timecode generators related to realtime have to reset once a day.
Maybe sometimes there will be a drop-drop timecode to upgrade from Julian to Gregorian calendar.
Evertz TC Generators :
I work with Evertz tools all the time, they are professional studio equipment. Good support also...
But for field use I prefer ambient products (Lockit system) because of portability and battery power supply.
The master clock system controller ACC 101 is a powerful tool for on set TC work. It generates, compares, reads TC in almost all formats I can imagine on a film/HD/video set. It also allows you to connect a GPS receiver (Garmin GPS mouse as an example) if you want to sync the system to GPS. DCF sync and sync to different framerates as the internal frame rate is also supported.
check http://www.ambient.de for details
Chris, GÃ¼nther and their crew also have a wireless system for TC and they support me since years.
Disclaimer: I donâ€™t work for ambient but Iâ€™m a really satisfied customer!
I never had problems with these units, the master clock runs about one week with a battery pack, only the lockits need more power because of the sync generator. If you use a unit like this, you donâ€™t forget the Ubits - better : youâ€™ll use them even more because itâ€™s so simple to change/set them and then just go to the lockits and jam them to the new settings. (Yes - I was asking for a wireless update of the Ubits, itâ€™s there but nobody else asked, so they donâ€™t finish it to a "real" product right now...) need more "askers".
Or: TriLvlSync on F900, word clock an DAT, two different frame rate on set - no problem!
Timecode is the only worldwide standard for sync on set and a powerful tool. If you use the genlock as well you are very accurate and wonâ€™t have problems in post.
AND : Time of day TC needs more attention - I agree absolutely...Userbits are more important. But I donâ€™t see a problem working with ToD, there are simple rules like changing tape every day (if you are shooting regular days like 8AM-8PM) and mark them; even if youâ€™re crossing midnight you just need to look for no double TC on a single tape...Just hire professional sound / DIT crews and it should work!
Even if you are recording rec run TC on LTC, just use one audio track for the audio-decks ToD TC and backup with a TC Slate...
So many ways to do it - just do it in a proper way!
+++ Florian Rettich +++
+++ Europe based DIT / vision control +++