gstreamer/docs/random/design
2011-09-02 19:26:43 +02:00

440 lines
15 KiB
Text

GStreamer Editing Services
--------------------------
This is a list of features and goals for the GStreamer Editing
Services.
Some features are already implemented, and some others not. When the
status is not specified, this means it is still not implemented but
might be investigated.
FUNDAMENTAL GOALS:
1) API must be easy to use for simple use-cases. Use and abuse
convenience methods.
2) API must allow as many use-cases as possible, not just the simple
ones.
FEATURES
Index of features:
* Project file load/save support (GESFormatter)
* Grouping/Linking of Multiple TrackObjects
* Selection support (Extension from Grouping/Linking)
* Effects support
* Source Material object
* Proxy support
* Editing modes (Ripple/Roll/Slip/Slide)
* Coherent handling of Content in different formats
* Video compositing and audio mixing
* Handling of alpha video (i.e. transparency)
* Faster/Tigher interaction with GNonLin elements
* Media Asset Management integration
* Templates
* Plugin system
* Project file load/save support (GESFormatter)
Status:
Implemented, requires API addition for all use-cases.
Problems:
Timelines can be stored in many different formats, we need to
ensure it is as easy/trivial as possible for users to load/save
those timelines.
Some timeline formats might have format-specific
sources/objects/effects which need to be handled in certain ways
and therefore provide their own classes.
The object that can save/load GESTimeline are Formatters.
Formatters can offer support for load-only/save-only formats.
There must be a list of well-known GES classes that all formatters
must be able to cope with. If a subclass of one of those classes is
present in a timeline, the formatter will do its best to store a
compatible information.
A Formatter can ask a pre-render of classes that it doesn't
understand (See Proxy section).
Formatters can provide subclasses of well-known GES classes when
filling in the timeline to offer format-specific features.
* Grouping/Linking of Multiple TrackObjects
Status:
Implemented, but doesn't have public API for controlling the
tracked objects or creating groups from TimelineObject(s)
Problems:
In order to make the usage of timelines at the Layer level as easy
as possible, we must be able to group any TrackObject together as
one TimelineObject.
The base GESTimelineObject keeps a reference to all the
GESTrackObjects it is controlling. It contains a mapping of the
position of those track objects relatively to the timeline objects.
TrackObjects will move and be modified synchronously with the
TimelineObject, and vice-versa.
TrackObjects can be 'unlocked' from the changes of its controlling
TimelineObject. In this case, it will not move and be modified
synchronously with the TimelineObject.
* Selection support (Extension from Grouping/Linking)
Problems:
In order to make user-interface faster to write, we must have a way
to create selections of user-selected TimelineObject(s) or
TrackObject(s) to move them together.
This should be able to do by creating a non-visible (maybe not even
inserted in the layer?) TimelineObject.
* Effects support
Status:
Partially Implemented, requires API addition for all use-cases.
Problems:
In order for users to apply multimedia effects in their timelines,
we need an API to search, add and control those effects.
We must be able to provide a list of effects available on the system
at runtime.
We must be able to configure effects through an API in GES without
having to access the GstElements properties directly.
We should also expose the GstElements contained in an effect so it
is possible for people to control their properties as they wish.
We must be able to implement and handle complex effects directly in
GES.
We must be able to configure effects through time -> Keyframes
without duplicating code from GStreamer.
* Source Material object
Problems:
Several TimelineSource for a same uri actually share a lot
in common. That information will mostly come from GstDiscoverer,
but could also contain extra information provided by 3rd party
modules.
The information regarding the various streams (and obtained through
optionally running GstDiscoverer) is not stored and has to be
re-analyzed else where.
Definition:
Material: n, The substance or substances out of which a thing is or
can be made.
In order to avoid duplicating that information in every single
TimelineSource, a 'Material' object needs to be made available.
A Material object contains all the information which is independent
of the usage of that material in a timeline.
A Material contains the list of 'streams' that can be provided with
as much information as possible (ex: contains audio and video
streams with full caps information, or better yet the output of
GstDiscoverer).
A Material contains the various Metadata (author, title, origin,
copyright ,....).
A Material object can specify the TimelineSource class to use in a
Layer.
* Proxy support
Problems:
A certain content might be impossible to edit on a certain setup
due to many reasons (too complex to decode in realtime, not in
digital format, not available locally, ...).
In order to be able to store/export timelines to some formats, one
might need to have to create a pre-render of some items of the
timeline, while retaining as much information as possible.
Content here is not limited to single materials, it could very well
be a complex combination of materials/effects like a timeline or a
collection of images.
To solve this problem, we need a notion of ProxyMaterial.
It is a subclass of Material and as such provides all the same
features as Material.
It should be made easy to create one from an existing TimelineSource
(and it's associated Material(s)), with specifiable rendering
settings and output location.
The user should have the possibility to switch from Proxy materials
to original (in order to use the lower
resolution/quality/... version for the editing phase and the
original material for final rendering phase).
Requires:
GESMaterial
* Editing modes (Ripple/Roll/Slip/Slide)
Status:
Not implemented.
Problems:
Most editing relies on heavy usage of 4 editing tools which editors
will require. Ripple/Roll happen on edit points (between two clips)
and Slip/Slide happen on a clip.
The Ripple tool allows you to modify the beginning/end of a clip
and move the neighbour accordingly. This will change the overall
timeline duration.
The Roll tool allows you to modify the position of an editing point
between two clips without modifying the inpoint of the first clip
nor the out-point of the second clip. This will not change the
overall timeline duration.
The Slip tool allows you to modify the in-point of a clip without
modifying it's duration or position in the timeline.
The Slide tool allows you to modify the position of a clip in a
timeline without modifying it's duration or it's in-point, but will
modify the out-point of the previous clip and in-point of the
following clip so as not to modify the overall timeline duration.
These tools can be used both on TimelineObjects and on
TrackObjects, we need to make sure that changes are propagated
properly.
* Coherent handling of Content in different formats
Problems:
When mixing content in different format (Aspect-Ratio, Size, color
depth, number of audio channels, ...), decisions need to be made on
whether to conform the material to a common format or not, and on
how to conform that material.
Conforming the material here means bringing it to a common format.
All the information regarding the contents we are handling are
stored in the various GESMaterial. The target format is also known
through the caps of the various GESTracks involved. The Material and
track output caps will allow us to make decisions on what course of
action to take.
By default content should be conformed to a good balance of speed
and avoid loss of information.
Ex: If mixing a 4:3 video and a 16:9 video with a target track
aspect ratio of 4:3, we will make the width of the two videos
be equal without distorting their respective aspect-ratios.
Requires:
GESMaterial
See also:
Video compositing and audio mixing
* Video compositing and audio mixing
Status:
Not implemented. The bare minimum to implement are the static
absolute property handling. Relative/variable properties and group
handling can be done once we know how to handle object grouping.
Problems:
Editing requires not only a linear combination of cuts and
sequences, but also mixing various content/effect at the same
time.
Audio and Video compositing/mixing requires having a set of base
properties for all sources that indicate their positioning in the
final composition.
Audio properties
* Volume
* Panning (or more generally positioning and up-/down-mixing for
multi-channel).
Video properties
* Z-layer (implicit through priority property)
* X,Y position
* Vertical and Horizontal scaling
* Global Alpha (see note below about alpha).
A big problem with compositing/mixing is handling positioning that
could change due to different input/output format AND avoiding any
quality loss.
Example 1 (video position and scale/aspect-ratio changes):
A user puts a 32x24 logo video at position 10,10 on a 1280x720
video. Later on the user decides to render the timeline to a
different resolution (like 1920x1080) or aspect ratio (4:3 instead
of 16:9).
The overlayed logo should stay at the same relative position
regardless of the output format.
Example 2 (video scaling):
A user decides to overlay a video logo which is originally a
320x240 video by scaling it down to 32x24 on top of a 1280x720
video. Later on the user decides to render a 1920x1080 version of
the timeline.
The resulting rendered 1920x1080 video shall have the overlay
video located at the exact relative position and using a 64x48
downscale of the original overlay video (i.e. avoiding a
640x480=>32x24=>64x48 double-scaling).
Example 3 (audio volume):
A user adjusts the commentary audio track and the soundtrack audio
track based on the volume of the various videos playing. Later on
the user wants to adjust the overall audio volume in order for the
final output to conform to a target RMS/peak volume.
The resulting relative volumes of each track should be the same
WITHOUT any extra loss of audio quality (i.e. avoiding a
downscale/upscale lossy volume conversion cycle).
Example 4 (audio positioning):
A user adjusts the relative panning/positioning of the commentary,
soundtrack and sequence for a 5.1 mixing. Later on he decides to
make a 7.1 and a stereo rendering.
The resulting relative positioning should be kept as much as
possible (left/right downmix and re-positioning for extra 2
channels in the case of 7.1 upmixing) WITHOUT any extra loss in
quality.
Create a new 'CompositingProperties' object for audio and video
which is an extensible set of properties for media-specific
positioning. This contains the properties mentionned above.
Add the CompositingProperties object to the base GESTrackObject
which points to the audio or video CompositingProperties
object (depending on what format that object is handling).
Provide convenience functions to retrieve and set the audio or video
compositing properties of a GESTrackObject. Do the same for the
GESTimelineObject, which proxies it to the relevant GESTrackObject.
Create a new GESTrack{Audio|Video}Compositing GstElement which will
be put in each track as a priority 0 expandable GnlOperation.
That object will be able to figure out which
mixing/scaling/conversion elements to use at any given time by
inspecting:
* The various GESTrackObject Compositing Properties
* The various GESTrackObject GESMaterial stream properties
* The GESTrack target output GstCaps
The properties values could be both set/stored as 'relative' values
and as 'absolute' values in order to handle any input/output formats
or setting.
Objects that are linked/grouped with others have their properties
move in sync with each other. (Ex: If an overlay logo is locked to a
video, it will scale/move/be-transparent in sync with the video on
which it is overlayed).
Objects that are not linked/grouped to other objects have their
properties move in sync with the target format. If the target format
changes, all object positioning will change relatively to that
format.
Requires:
GESMaterial
See also:
Coherent handling of Content in different formats
* Handling of alpha video (i.e. transparency)
Problem:
Some streams will contain partial transparency (overlay
logos/videos, bluescreen, ...).
Those streams need to be handle-able by the user just like
non-alpha videos without losing the transparency regions (i.e. it
should be properly blended with the underlying regions).
* Faster/Tighter interaction with GNonLin elements
Problems:
A lot of properties/concepts need to be duplicated at the GES level
since the only way to communicate with the GNonLin elements is
through publically available APIs (GObject and GStreamer APIs).
The GESTrackObject for example has to duplicate exactly the same
properties as GnlObject for no reason.
Other properties are also expensive to re-compute and also become
non-MT-safe (like computing the exact 'tree' of objects at a
certain position in a GnlComposition).
Merge the GES and GNonLin modules together into one single module,
and keep the same previous API for both for backward compatibility.
Add additional APIs to GNonLin which GES can use.
* Media Asset Management integration
(Track, Search, Browse, Push content) TBD
* Templates
Problem:
In order to create as quickly as possible professional-looking
timelines, we need to provide a way to create 'templates' which
users can select and have an automatic timeline 'look' used.
This will allow users to be able to quickly add their clips, set
titles and have a timeline with a professional look. This is
similar to the same feature that iMovie offers both on desktop and
iOS.
* Plugin system
Problem:
All of GES classes are made in such a way that creating new
sources, effects, templates, formatters, etc... can be easily added
either to the GES codebase itself or to applications.
But in order to provide more features without depending on GES
releases, limit those features to a single application, and in
order to provide 'closed'/3rd party features, we need to implement
a plugin system so one can add new features.
Use a registry system similar to GStreamer.