OVERVIEW
NEWS
OBJECTIVES
PARTNERS
STATE OF THE ART
RESEARCH AREAS
INNOVATION
IMPACT
RESEARCH PLAN
DOCUMENTS
LINKS
CONTACTS
EVENTS
MOVIES
FREE

 State of the art
Integrated Project Research Area Cinema

The most advanced aspect of the Digital Cinema process in 2007 is the ‘Digital Intermediate’ stage which links the stages from ‘shot’ to ‘screen,’ forming an essential prerequisite for all Digital Film/D-Cinema/E-Cinema scenarios and cross-platform exploitation. The Digital Intermediate is a file of the whole movie, which brings together all the original material (analogue film, digital camera, or CGI), which can then be processed and postproduced in a ‘digital lab.’

After ‘finishing,’ the digital intermediate becomes the digital film master, from which distribution copies can be made in the appropriate resolution, format and medium (analogue film, digital cinema playout, DVD, TV et cetera). The digital intermediate process was pioneered in 1999, when the Computer Film Company in London digitally scanned, graded, conformed and recorded Chicken Run in its entirety. By 2006, about 50% of Hollywood production used a digital intermediate process, and this is likely to rise to over 70% of Hollywood and European features by 2008.

Digital Intermediate files are uncompressed (to ensure maximum quality) with a colour bit depth of 10-bit log per channel RGB, or up to 16 bit linear. Most Digital Intermediates are ‘2K’ with a frame resolution of 2048 x 1556 pixels, but there is a growing industry demand for 4K. The size of a Digital Intermediate file (typically approaching 2 Terabyte for 2K and 8 Terabyte for 4K) imposes heavy requirements for storage, data processing and management.

Current suggestions for digital laboratories suffer from the problems of storing and moving very large data volumes between multiple workstations, which outstrip the capacity of the current standard data networking methods used in the industry (Gigabit Ethernet, SAN or HSDL streaming). Gigabit Ethernet can transfer 2K movie files (12 MB of data per frame) at data frames at around 6 fps, a SAN at up to 14fps and HSDL at 15 fps. Many Postproduction operations involve the movement and combination of multiple streams and the increasing demand for 4K resolution makes the problem four times worse, with a requirement to move 48 MB of data per frame. The desire to work not just with one movie but to hold many movies and their rushes ‘live on line’ further magnifies the data management problem.

In current practice, digital acquisition, postproduction and display are independent of one another, and the inclusion of one process in a project doesn't immediately require the involvement of the other two: digitally acquired material can just as easily be projected via traditional celluloid means as via a digital projector and film originated material may be shown digitally. The Digital Intermediate process is therefore applicable to traditional film projects as well as those classified using digital capture or projection techniques.

Initial digital content may be captured by digital cameras; scanned from analogue film; or wholly synthesised as CGI. Digital cameras are at an early stage of development and cannot yet match analogue film in either spatial or colour resolution. Nevertheless, a growing number of directors are adopting digital cameras. But as long as cinematographers prefer analogue cameras, the main route from film to digital intermediate will be via either film scanner or data Telecine.

Considerable research is under way into creating autonomous synthetic actors, but this is some way from film industry applications. Final Fantasy (2001) received considerable attention as the first ‘all CGI’ movie, aiming at photo-realism, with a connection to the games market. But the cost of modelling and animating not-quite-realistic synthetic actors remains prohibitively high. The most effective examples are those where the character is not quite human (e.g. ‘Gollum’ in Lord of the Rings 2 and 3).

Particle animation and behavioural animation are now regularly used to create complex battle scenes and crowds. Integration of real and virtual sets and characters poses severe problems for actors as well as filmmakers, who require real-time visualisation to see what is going on. Real-time virtual studios currently run at TV resolution, either with low realism or with high realism (using an approximation of ray tracing) but a fixed camera viewpoint. More exotic techniques, involving surface capture with 3-D motion scanners creating 3-D/4D photo-real model data are at the laboratory stage.

Postproduction processes divide between Digital Intermediate (DI) ‘bulk’ operations and visual effects applied to short scenes (such as explosions). While there is a trend toward software solutions and standard platforms (such as Linux-based clusters), the need to perform operations on very large files currently requires special-purpose hardware for real-time DI processes such as scaling, compositing, grading and colour correcting. Devices such as the DVS CLIPSTER are used to store, edit and  process uncompressed film data in resolutions of more than 2.000 pixels per line in real time without rendering, for controlling and executing Digital Intermediate operations. 

They also integrates with other high-end devices, such as the Pandora MegaDef for real-time colour correction.. Several FP5 projects have been carrying out RTD to extend the scope and reduce the cost of DI work. RACINE-S project (IST-2001-37117) addresses working on novel methods for the regeneration of missing frames by the use of 2.5D and 3-D techniques. The Speed-FX project (IST-2001-34337) proposes a low-cost architecture, based on standard components, to run full-resolution DI processes for colour grading and a range of effects. All these approaches use standard architectural elements, to increase programmability and plug-in development, with specialised hardware components (such as FPGAs and novel i/o structures) to obtain real-time working. Complex visual effects – such as explosions – may take weeks to create and involve numerous techniques and platforms for 3-D modelling, animation, rendering, compositing and colour correction.

The digital production chain encourages the adoption of standardised metadata formats and file wrappers such as AAF and MXF for communication between the (often proprietary) file formats that proliferate in the film industry. In theory, the DI file and new processing platforms should make it possible to automate many processes along the chain, using techniques such as object tracking and sound or image-based information retrieval: in practice, these techniques are in their infancy.

Files for digital distribution and projection (as opposed to output back to film for analogue projection) are transported to the viewing theatre by satellite, dedicated landline or encrypted data cartridge. The viewing file is at a lower resolution than the Digital Intermediate, compression is permitted, and data security is at a premium. The theatre stores the file in a specialised server, which controls decryption and playout. This can provide version control, automated programming for advertisements, multilingual subtitling, and reliable audit trails as well as the multi-projection and theatre functions of a multiplex.
By mid-2007, about 400 European cinemas (or only 1.3% of the total) were equipped for high-quality ‘D-Cinema’ digital projection – although this number is growing fast and many more venues are equipped with lower-quality ‘E-Cinema’ digital projection facilities. The projector technology is usually based on the Texas Instruments Digital Light Processing (DLP) chip, which is licensed by specialised manufacturers.

 
 NOTES

The team responsible for creating CFC’s Digital Film Lab technology now comprises a separate company, FilmLight, which is a partner in IP-RACINE.

It is argued that at least14 bits (16384 levels) is necessary to represent the full dynamic range of film, but log data provides a better representation of the way in which film (as opposed to a CCD) registers light, so that 10 bit log may do the job and linear files will contain massive redundancy for scanned data.  The two currently used standard formats are Cineon DPX (10-bit log) and .CIN (16 bit linear).

Digital image resolutions vary from around 720 x 578 pixels for Standard Definition Television, through 1920 x 1080 for High Definition Television, to 2048 x 1556 and 4096 x 3172 pixels for 2K and 4K film resolution. Full aperture, 35mm Academy format film gives an image 24.576 x 18.672mm. 4K line resolution (4096 x 3112 square pixels) and is equivalent to a 6µm resolution, nominal 2K equivalent to 12µm.

For example, George Lucas Star Wars Episode 2 was shot on the CineAlta in HD 1920 x 1080, with no noticeable loss of quality and a reported cost saving of $2 million. In this case, frames were so composited with digitally-generated backgrounds that the camera material was only a part of many images.

Telecine machines were invented to scan cine film in ‘real time’ and produces electrical signals for either television broadcast purposes or for recording onto tape or disk. These machines have been known since the 1920s and date back to John Logie Baird, the inventor of television. Before the days of reliable video recording systems, films were broadcast directly from Telecines. Nowadays, ‘state of the art’ Telecines are capable of real time standard definition transfers  and real time HDTV transfers. Some Telecine systems offer higher resolution scanning capabilities – though not necessarily in real time. The Thomson Spirit 4K is designed to scan 2K in real time, and 4K in non-real-time. Because of the transport scheme used to move film at 24 or 25 frames per second, the film drive is not usually sprocket driven. Instead, ‘edge guidance’ of film is used.

Film Scanners have been known for many years. The essential difference used to be that a film scanner never worked in real time. However with ‘non-real time’ Telecines, this distinction has blurred. Typically film scanners work at 2000 or 4000-line resolution, and scan at between 4 seconds per frame and 4 frames per second. Film scanners nearly always use sprocket drives and ‘pin registration’ for transport and positioning, to give very accurate position and stop the image ‘wavering.’

Performance differences between Telecine and film scanners are typically in ‘repeatability’. Because of the pin registration on film scanners, the accuracy of the scanning and the precision of the scanning raster are far greater than on Telecine machines. Telecine machines usually grade ‘on the fly’ and have large amounts of real time hardware to process the ‘look’ of the pictures before recording or storing the digital representation. Film scanners have far less capability for this sort of processing, which is normally done later (i.e. after the image has been recorded or stored).

Final Fantasy was modelled and animated using Maya, with extensive purpose-written extensions for clothes and hair, and rendered with Renderman. To give an idea of the scale, the final output render took 934,162 CPU-days of render time on a render farm with approximately 1,200 processors for a movie with 150,000 frames.  Including test renders, revisions, and reviews, the render farm software ran about 300,000 jobs, with an average of 50-100 frames/job.  See here for details.

By 3-D/4D we refer to the use of dynamic 3-D scanners to capture complete 3-D data about a scene, together with temporal data. This would make it theoretically possible to navigate and modify scenes in four dimensions, treating time in the same way as the three dimensions of space. This is currently a theoretical possibility, but the Matrix movies show that it might have practical interest for filmmakers at some future date. See 3-D-Matic laboratory, Faraday Imaging Partnership, http://www.faraday.gla.ac.uk/

‘Colour corrector’ is actually a misnomer for a device that, in addition to carrying out real-time colour correction, executes a variety of different transformations by manipulating the colour of pixels in the data stream.

The Advanced Authoring Format AAF is a software implementation of SMPTE metadata and labels, designed to make it easy to work with large collections of inter-related sets of metadata and essence. The AAF Association is primarily made up of broadcasters and companies with digital media interests. It works closely with the SMPTE and EBU.

The Material Exchange Format (MXF) was originated by the Pro-MPEG forum to handle digital-file exchange for audio and video files (compressed or uncompressed). An MXF file serves as a data wrapper that can contain any type of audio or video material in a playable format. MXF is designed for compatibility with AAF and is supported by the G-FORS (Generic Format for Storage) project IST-1999-11238. MXF is being proposed to the SMPTE as a formal standard.

Digital projection is also growing in postproduction houses and studios, as the best way of viewing DI.

This essentially employs arrays of steerable micro-mirrors, one per pixel, mounted in a rectangular array in an optical semiconductor known as the Digital Micromirror Device, or DMD chip, which was invented by Dr. Larry Hornbeck of Texas Instruments in 1987. The DMD chips in the DP50 1280 x 1024 projector have 1.3 million hinge-mounted micro-mirrors and one chip is used for each of the RGB channels. The Barco DP-2000, with a 2K resolution is one of the current market leader, providing a viewing experience better than any analogue projector. The principal rival technologies (from Sony and JVC) are less bright but proposes a 4K projector for evaluation.  The industry goal is now to drive down cost, improve brilliance, and expand the experience into 3-D viewing.