Update on the Interoperable Mastering Format

April 5, 2014, NAB, Las Vegas—A panel of content owners looked at the use of interoperability standards in their work. Howard Lukk from Disney moderated the panel, while panelists were Annie Chung from Disney, Arjun Ramamurthy from 20th Century Fox, and Bill Baggelaar from Sony.

The interoperability mastering format is a new and improved implementation of the original DCI (digital cinema interoperability) spec. After 14 years, the spec is still unused by the industry, so this new IMF spec has been developed through work at the ETC at USC. The IMF is based on some of the SMPTE DCI documents and is slowly being rolled out.

Chang, who is also the IMF working group chair, described the basic functions for the IMF. The original efforts in the DCI were to help the non-digital theaters make the transition. The latest IMF versions allow for use of a partial package of functions rather than calling for full implementation. The IMF can store versions as instances, allowing a true file-based workflow.

While based on the SMPTE DCI standard, the IMF can allow segmented files and versions by defining an output profiles list and a sample materials interchange. The IMF composition tracks all the included files and the output is a playlist of audio, video, clusters, and metadata. This list tracks the various files, outputs the profile list, manages assets, and puts them all together in a packing list.

The standard allows for modular implementation by supplying a framework and links to the necessary applications needed to configure and process all of the input files. The latest specs are available as published documents as well as lists of other work in progress.

Ramamurthy noted that the IMF packages act as a collection of assets and recipes to crate the exponentially exploding versions. As a result, the IMF supports the whole digital media supply chain, international distribution, and monetization. The IMF can generate a localized version from the master domestic files via some metadata. The actual process of tracking the files doesn’t change between theatrical and episodic versions.

This differs from the current process of localizing the content as xml files. Now catalogs can reference localized versions allowing for remote management software to track the releases. One flow is to ingest the domestic flow and a text-less set of files. The localized source components are compressed and mapped to a reference for synchronization due to the differences in run times for the ads, and other local changes. The still have to match the domestic source with the local one.

Next they generate the local text information, identify the text information in the domestic version, and match and clip the international and domestic versions for sync purposes. They can abstract the information into the playlist and generate the new local version from a playlist.

The components for the conversions include the domestic content with text-less files, localized content, localized text, and a localized xml playlist. The asset management comes from linkages and provides workflow orchestration. These files are integrated with normal non-linear editing and importing and exporting appropriate graphics. The standard NLE and a message layer wrapper form a state machine.

The ingest, edit, and text operations are part of the overall orchestration and form an asset management system for the flows to transcode the assets for delivery. The output file list specifies all of the transcoding functions in xml. Changes in the content play list and the output pay list are addressed by adding processes.

Chang continued that the IMF at Disney and ABC have started testing and first implementations. File-based workflows started in ’05 with “Cars” from Pixar, leading to work with ETC at USC in ’07. This work on the standards proceeded through ’11 where more formats and distribution functions emerged. The IMF is needed because a film can generate 35k files. This year, they converted to IMF to handle the mass of files.

Now, the titles and other text are more easily changed for different versions. They only have to store the changed pieces and identify them in a process file. Now the localized versions are only different from the domestic version in text and audio for speech. As a result, they reduced their storage needs by a large amount because the movie is only the domestic version plus local speech and text rather than a full version for each locality.

One clean-up project is to move all assets into the Disney Anywhere system that is a master asset collection plus trailers and artwork. The entirety is part of a master asset package that includes collections, maps of the files, tracks, home extras, that are described through metadata. The conversions are starting this year.

These changes are also going to the ABC side, since the IMF is not just for cinema, but is applicable for global self services. Now shows have three phases, English only, non-English supplements, and the flow for interchange and creation for downstream delivery. The system also facilitates storing and archiving for those files. The IMF enables other functions like an intelligent library, marketing and promo versions, and greater integration of other flows into the IMF. Given all this effort, they still need more tools for edit, package creation, transcoding, and color correction.

Baggilaar commented that they are working to deliver UHD. One goal was to generate 100 UHD titles within a year. This work called for remastering and delivering existing movies in UHD. The project to support 4k displays requires remastering and changing color depth in the existing films. The workflow supports masters in 4k Blu-Ray from IMF and mates with the existing workflows.

They go back to the original film for the masters and reassemble and recolor them for processing. This flow generates from 800TB to 1PB for all of the files for a film. Some issues they have to address are sync, 4k delivery, and the long-term value versus the needs of the project of converting to the IMF. They are working in a 3840 X 2160 format encoded into a YCbCr format. The conversion process generates an average of 400 MB/s with a peak of 800 MB/s.

The workflow takes a master and applies color balance and corrections, then converts the files to JPEG 2000. a UHD wrapper is added and all are moved to a components play list for movement to the IMF. The component-based nature of IMF generates a total package of 182 GB, with a CPL of 168KB for descriptors. They converted 280 films into UHD, each about 2 hours of 4k content. The package and component nature of the IMF allowed the 1PB to be compressed to 43TB.

The workflow is also being applied to new releases to give more localization while allowing more time to prepare the files. The IMF allows them to generate all aspect ratios and localization from a 4k TV master. The localization using a component play list allows for changing sections of a film and the tech files can be a 40GB base file plus 10GB for a language. The benefits of the reduced data loads helps to keep the files in sync and still retain the quality of the original. Netflix is doing everything in IMF.