Entries Tagged 'Transmageddon' ↓

Update from GStreamer Hackfest at Google Office in Munich

To give the wider community a chance to see what happened during the GStreamer hackfest last weekend I put together this blog post is based on an summary written by Wim Taymans, so a big thanks to Wim for letting me reuse parts of his summary.)

So last weekend 21 GStreamer hackers got together at the Google office in Munich to spend the weekend hacking on their favourite GStreamer bits. At this point in time we didn’t have any major basic plumbing tasks that needed tackling so the time was spent hacking on a lot of different projects and using the opportunity to discuss design and challenges with each other.

We where 3 people attending from Red Hat and Fedora; Wim Taymans, Alberto Ruiz and myself.

With the Release of GStreamer 1.0 in September 2012, we drastically changed the way memory is handled in the multimedia pipeline and the large body of work is still in exploring, improving and porting elements to this new memory model. We’re also mostly working on improving the existing elements with comparatively little new infrastructure work.

We’re also seeing a lot of people from different companies that contribute significant amounts of code to the official GStreamer repositories. This has traditionally been a much more closed effort with various pieces of code living in multiple repositories, especially for the hardware acceleration bits. It is good to see that the 1.0 series brings all these efforts together again with more coordination and a more coherent story.

HW acceleration

One of the large ongoing tasks is to improve our support for hardware accelerated decoding, effects and display. With 1.0 we can finally get this done cleanly and efficiently in very many use cases.

Matthew Waters to flew in from Australia to move the gst-plugins-gl set of plugins to the core GStreamer plugins packages. He has been working on these plugins for a while now. Their goal is to use OpenGL to apply operations on the video, like rotation on a cube or applying a shader. With the 1.0 memory management it becomes possible to do this efficiently with a minimal amount of texture upload/downloads. More work is needed here, we can optimize things some more by delaying the work and running the shaders as part of the rendering operation.

Andoni Morales (Fluendo) has also been working on improving hardware acceleration on android. He used some of the new features of 1.0 to make the android codecs use zero-copy by implementing the texture-upload meta data on buffers. This allows the video sink to efficiently create a texture from the decoded data for display. Andoni also ported winks, a video capture source on Windows, to GStreamer 1.0.

Nicolas Dufresne (Collabora) has been working on adding a new set of decoders based on the mem2mem API in v4l2. Not many drivers provide this API yet but it is implemented in some Samsung Exynos SOCs. We would also like to support other m2m operations later, such as color conversion but for that we need to make some of our base classes support the required asynchronous behaviour of mem2mem. The memory management in our v4l2 elements has been gone through several iterations of improvements during the 1.0 cycle but it still is not entirely what it should be. We agreed on what we should do to fix this in the near future. We also briefly discussed the need for a new event that can be used to reclaim memory from a pipeline; many elements that use hardware buffers need to free those before they can negotiate a new format with the hardware so we need a way to make that possible.

Mathieu Bourron (Collabora) has been working on libva, the library for GPU based video decoding and encoding on Intel hardware, and spent his time at the Hackfest fixing up the SPU overlay element to enable hardware accelerated subpicture overlays in the video sink. Traditionally GStreamer would use the CPU to overlay the subpictures (of DVD, for example) on top of the video images. With new GL-based sinks, and hardware accelerated decoders this is very undesirable and can be done much more efficient as part of the final rendering. In 1.0 we have the infrastructure to delay this overlay operation by attaching extra metadata (with the subpicture) to the video images when the video sink knows how to overlay them. We have been doing this with subtitles in cluttersink and other sinks for a while now and soon we can also do this with subpictures.

Plugin Hacking

Arun Raghavan, GStreamer hacker and Pulse Audio maintainer, worked on disabling the audio and video filters in playbin when passthrough mode was selected. In passthrough mode, a video or audio sink can directly handle the encoded media (think a bluetooth headset that can handle mp3 directly or a hardware sink that takes encoded data). He expaned on that work in blog entry.

As a cool hack, Arun also made a source element to read from torrent files so you can watch a movie while you torrent it. He provides more information on that element in his blog, it is actually really cool.

Thiago Santos (Collabora) was continuing with his work to improve the DASH demuxer, reworking the buffering code to make it buffer less and smoother. Dash is one of the new formats (with HLS and MSS) to stream media over HTTP while adapting to bandwidth changes. On the server side, it makes media available in various bitrates while a client switches between bitrates depending on its measured network conditions. Andoni Morales also worked on a new dashsink element that implements the server side of the DASH
format.

Mathieu Duponchelle, a former GSoC student was trying to improve support for seeking in MPEG Transport Streams in order to use them in PiTiVi. Seeking in MPEG TS is not an easy thing because they are really optimized for streaming only. He got help from Thibault Saunier (Collabora), who was also hacking on PiTiVi and who was preparing a new release of gnonlin, GES and gst-python 1.2 (which he released on Sunday). Mathieu is one of the developers able to work fulltime on PiTiVi now thanks to the PiTiVi fundraiser, so be sure to contribute to that!

Jan Schmidt (Centricular), a long time GStreamer core hacker was working on debugging some DVB issues and also ended up taking part in a lot of the general design and troubleshooting discussions happening during the hackfest, helping other people move forward with their projects.

Long time GStreamer hacker Edward Hervey (Collabora) was planning to do a lot of DVB hacking but had to give up on that effort when it was clear that Google had signal isolated the office for security reasons, so there was no DVB signal in the Google office. Instead he worked on merging some pending DVB patches and implemented GAP support in the mpeg transport stream plugin. GAP support deals with streams that have long periods of no media (like missing audio for some time in DVD). It makes sure that downstream elements keeps processing the silence instead of waiting for more data.

Applications

Meg Ford, a GSOC student mentored by Sebastian Dröge (Centricular) was working on Gnome Sound Recorder and fixing up the last bugs, preparing it for a new release.

Myself, Christian Schaller (Red Hat) was on a bug fixing spree in Transmageddon (a transcoding application written in python and GStreamer) and managed to reduce the number of known bugs to only 1. Fixed that last bug once I got home, so now I just need to hammer at Transmageddon for a bit to make sure I caught all the corner case issues so I can do a major new release with new features such as handling files with multiple audio streams, handling DVD ripping, handling VP9 encoding, handling setting audio stream language information, reducing decoding overhead for streams that we are going to throw away and more. Also had help reviewing and cleaning up the Transmageddon code from Alberto Ruiz, freeing Transmageddon from some ugly code that had survived many library updates and rewrites.

Alessandro Decina(Spotify) kept working on his patches to update the Firefox GStreamer backend to GStreamer 1.0. We hope to deploy this work in Fedora in the not to distant future. As a hack for the hackfest he provided patches to implement audio and video capture.

Wim Taymans (Red Hat) was hacking on a new library that can parse and generate MIKEY messages (RFC 3830). He want to use this in the GStreamer RTSP server to negotiate SRTP (secure RTP) encryption parameters.

We had 2 people from the Swedish company AXIS, who provide network cameras that all run GStreamer and who contribute on a regular basis to the RTP and RTSP elements and libraries. Ognyan Tonchev was mostly writing some unit tests for RTSP and multicast handling in the RTSP server. Sebastian Rasmussen had been hacking on our watchdog element and the payloaders.

Infrastructure

Long time GStreamer hacker Stefan Sauer (Google) gave a demo of his idea for a tracing infrastructure in GStreamer. The idea is to place trace macros at strategic places that would send structured data to pluggable tracer modules. Some of the tracer modules could, for example measure CPU usage of a plugin or measure the latency. The idea would be to gradually replace our extensive (but unstructured) logging with this new trace infrastructure. This would allow us to do new interesting things, like send debug log to a remote machine or produce STF (Structured Trace Format) to analyse with standard tools. No immediate plans were made to merge this but there seems to be very little resistance to get this merged soon.

Core hacker Sebastian Dröge (Centricular) has been going over the current Stream selection ideas. One of the long outstanding issues is that of switching streams between different languages: you have a movie in different languages and you want to switch between them. To achieve low-latency old data should be kept around for the streams that are not currently selected and be quickly and sent to the audio device. The idea is a combination of events to select a stream and to have the demuxer seek back in the stream on
switches. No final conclusion or plan that can solve all requirements has been reached yet.

Also investigations have begun to make decodebin deal with renegotiations. For example, when a new stream is selected, we might need to use a different decoder for this stream but also when new input is received, decodebin should be able to reconfigure itself. The decodebin code is a complicated beast so any change to it should be done carefully.

GStreamer maintainer Tim-Philipp Müller (Centricular) spent his time merging the new device probing and monitoring API (written by Olivier Crête from Collabora) that had been sitting in bugzilla for a while now. The purpose is to be able to probe devices and their capabilities such as v4l2 and ALSA devices. It’s also possible to be notified when devices appear and disappear in the system. An implementation for pulseaudio devices and another for v4l2 devices using gudev has been committed as well. This reimplements a
feature that was in 0.10, but got cut from 1.0 due to us not being happy with the old design. One of the complications with that was the fact that we ran out of bits in one of our enums so we needed to find a good solution for that.

We briefly discussed how to implement the SKIP seek flag. This extra flag can be used when doing fast forward or reverse and instructs the decoders that it is allowed to throw away data to more efficiently perform the trick mode (at reduced accuracy). There was a prototype for AVI playback that I implemented once that we discussed a bit. We’ll see if someone takes up the task to finalize this work and implement SKIP mode in more demuxers.

I took some photos during the event to capture the spirit and put them on Google Plus for your viewing pleasure.

A big thank you to Google for hosting us and providing us with free lunch and free drinks through the weekend.

PiTiVi fundraiser pass the 10 000 Euro mark

Hi, so I wrote A blog entry asking people to contribute to the PiTiVi fundraiser effort last week. I am happy to see them having already reached 10K Euro, but their goal of course is to get more. I am also happy to say that the GStreamer Project decided to support the fundraiser using some of the money we earned over the years doing Google Summer of Code and organizing the GStreamer Conference, which added 2500€ to the effort. You can read about the GStreamer donation here.

Anyway, I would once again want to ask people to contribute to this effort. There is a proven team behind this fundraiser, so this is not like a kickstarter where people are starting from scratch and you have no idea where they will end up going or what they might achieve. This is an existing project that just needs some polish to get it to critical mass. So visit the fundraising page and make your pledge.

PiTiVi Fundraising Campaign – Why you should donate

So the PiTiVi announced the PiTiVi fundraising campaign on Friday. I sincerely hope they are successful, because I think we really need a good non-linear video editor that runs on the Linux desktop, especially one that is built on top of GStreamer and thus sharing the core multimedia infrastructure of the rest of your desktop. The current PiTiVi team got the right skills and enthusiams in my view to truly pull this off, and their project is scoped in a manner that makes me believe they can pull it off. PiTiVi is already functional and this fundraiser is more about accelerating ongoing development as opposed to creating something from scratch. And their funding requirements for reaching the base milestone is rather modest, for example if just the employees of the 3 main linux distribution companies pitched in 3-4 Euro it would be enough to cover the base funding goal.

But I think this fundraiser is important also beyond the PiTiVi project, because it can serve as a precedent that it is possible to do significant crowdfunding around open source development and thus open the gate for more projects accelerating their development using it. There are a lot of great open source projects out there created on a volunteer basis, which is great and like PiTiVi they will flourish even without crowdfunding, but crowdfunding can be a great way for developers of the most interesting projects to be able to focus solely on their project for some time and thus accelerate its development significantly. So in the case of PiTiVi, I am sure the team will be able to achieve all the goals they have outlined in the funding campaign even if the fundraiser raises no money, but the difference here is if they do it in 1 year or 5 years.

So personally I donated 60 Euro to the PiTiVi fundraiser and I hope everyone reading this blog entry will do the same. Lets give the people developing this and other great open source tools our support and help them make their great software even better. This fundraiser is done by people passionate about open source and their project, because to be fair, no matter if the efforts ends up raising closer to 30 000€ or closer to 100 000 € it is in no way what anyone can call a get rich quick scheme, but rather modest amounts that will let two talented open source developers spend time working fulltime on a project we all want.

And remember whenever a major project using GStreamer gets a boost, it gives all GStreamer projects a boost. For instance in my own pet project, Transmageddon, I am gotten a lot of help over the years from general improvements in GStreamer done due to the involvement from PiTiVi developers, and I have even ended up copying code from PiTiVi itself a few times to quickly and easily solve some challenges in had in Transmageddon.

Hardware encoding in Transmageddon

So thanks to the new GStreamer 1.x VAAPI package for Fedora I was able to do a hardware encode with Transmageddon for the first time today. This has been working in theory for a while, but due to me migrating Transmageddon from GStreamer 0.10.x to GStreamer 1.x at the wrong time compared to the gstreamer-vaapi development timeline I wasn’t able to test it before.

Bellow is the GStreamer pipeline that Transmageddon use now if you have the Intel hardware encoder packages installed. Click on the image for the full image.
libva-pipeline-extract

I am also close to being able to release the new version of Transmageddon now. Most recent bugs found have turned out to be in various GStreamer elements, so I am starting to feel confident about the current pipeline building code in Transmageddon. I think that during the GStreamer Hackfest in Munich next Month I will make the new release with the nice new features such as general support for multiple audio streams, DVD ripping support and language tag setting support.

transmageddon-git-master

GStreamer Conference 2013 – Haggis edition

So we are quickly approaching the time for GStreamer Conference 2013. This year we will be in Edinburgh, Scotland and the conference will be hosted at the Edinburgh International Conference Centre alongside the Embedded Linux Conference Europe and the Automotive Linux Summit.

The GStreamer Conference 2013 Schedule is now live and as you can see there are a lot of great talks this year too, ranging from OpenGL integration, embedded hardware, new codecs and more. As always the GStreamer Conference is the best place to be to discuss the challenges of multimedia.

The conference will be held on the 22nd and 23rd of October so I strongly recommend you get yourself registered. If you want to attend ELCE or the Autotmotive summit make sure to register for those too and extend your stay in Edinburgh to also cover the 24th and 25th of October.

Looking forward to seeing you all there.

And last but not least, a big thank you to this years sponsors of the GStreamer Conference, especially Platinum sponsor Collabora. Also a big thank you to sponsors Fluendo, Google and for the first time this year Cisco.

And as always a big thank you to Ubicast who as always will be recording and publishing the conference for us. Be sure to check out their recordings from earlier years to find out what the GStreamer Conference is about.

GUADEC, Wayland, Transmageddon and more

So GUADEC is kicking off on Thursday here in Brno. The upcoming event is creating quite a bit of excitement here in the office as many members of the Red Hat team here in Brno has been helping out with the organization of the conference this year, being in the hometown of our biggest engineering office in the world. A series of last minute meetings, calls and arriving banners and packages help create a good buzz ahead of the opening of the conference. We have managed to get a bigger contingent of the Red Hat Desktop team this year than usual, including some members of our X.org/Wayland team, our Spice team and our LibreOffice team, so it will be a great opportunity for our global team to meet face to face in addition to meeting the other members of the community.

One of the items I am looking the most forward to during this GUADEC to is to be able to talk our friends at Intel and figure out how we can effectively work together on Wayland. We have put together a team to accelerate Wayland development inside Red Hat, with the goal of getting it ready for deployment in Fedora. There are of course a lot of things that needs to happen in preparation for this, like getting the GNOME Shell ready to work as a Wayland compositor. We will be looking at a long range of items, but among the more important ones is dealing with input devices in Wayland, like handling multiple mice, touch or Wacom tablets and making sure XWayland works so well that our users will not need to realize they are using Wayland instead of X once we transition over.

Anyway, I hope to share more details on our Wayland plans in the coming weeks and Months, so stay tuned.

I have also been trying to find some spare moments during the last few weeks so that I could celebrate GUADEC here in Brno with a new release of Transmageddon. There are basically three new features I am trying to polish up currently. Handling of files with multiple audiostreams is the most important, with DVD ripping support coming in as a close second and finally being able to tag the language of the audio streams as the third one. As you might imagine the first is also a pre-requisite for the second. I am getting there, although there are still some heavy lifting to do in terms of dealing with multiple audio streams inside Transmageddon in the case where the output doesn’t support multiple audio streams, like in the case of the FLV container format or if you are just stripping off the video to make a mp3 file.

Apart from that I think git master is starting to feel like working code again, I just need to test the hell out of it to catch all regressions my code re-organisation has caused. The new features feel small in some ways, but they took an awful lot of re-factoring as they obliterated some assumptions my code made. I am also quite happy to have managed to hide these new features pretty well for people who don’t need them, so if you are just loading a normal file without multiple audio tracks the UI should look and behave almost exactly as it did before. The same is true if you don’t have a DVD player on your system, you will not get any DVD items in the UI. I still need to make the code check for the availability of the libraries it needs for DVD support since they will not be shipping by default with most distros, and thus I will check for them and only make the DVD handling features available in the UI if they are installed.

So looking forward to seeing you here in Brno in the coming days!!

Brno GUADEC Call for Papers!

I would like to give everyone a friendly reminder that Saturday the 27th of April is the official deadline for the GUADEC 2013 Call for Papers. So make sure to get your proposal submitted.

We hope to have a wide range of talks this year, including talks on related subjects such as Wayland and Multimedia, so don’t automatically assume that you will not get a talk approved because its not ‘pure GNOME’.

GUADEC this year will be in Brno in the Czech Republic, so I hope to see as many of you as possible here.

GStreamer Hackfest in Milan

As those of you following the GStreamer development mailing list or the GStreamer Google Plus profile know, we have been having a GStreamer hackfest in Milan over the last few days. We have 17 people here, all hammering away at our laptops or discussing various technical challenges sitting at a nice place called the Milan Hub.

A lot of progress has been made during these days with some highlights including work on fixing the use of Gnonlin with GStreamer 1.0, which is a prerequisite for getting PiTiVi and Jokosher running with GStreamer 1.0. Jeff Fortin, Thibault Saunier, Nicolas Dufresne, Edward Hervey, Peteris Krishanis and Emanuele Aina has all been helping out with this in addition to fixing various other issues in PiTiVi and Jokosher.

Sebastian Dröge has put a lot of work during the hackfest into providing the basic building blocks for doing hardware codecs nicely in GStreamer, and Víctor Jáquez has been working on making VAAPI work well using these building blocks, with the plan among other things to make sure you have hardware accelerated decoding working with WebKit. In that regards Philippe Normand has spent the hackfest investigating and improving various bits of the GStreamer backend in Webkit, like improving the on-disk buffering method used. Also in terms of hardware codec support Edward Hervey also found a bit of time to work a little on the VDPAU plugins.

Speaking of web browsers Alessandro Decina has been working on porting Firefox to GStreamer 1.0, he has also been our local host making sure we have found places to eat lunch and dinner that where able to host our big group. So a big thank you to Alessandro for this.

Wim Taymans has been working on properly dealing with chroma keying in GStreamer, improving picture quality significantly in some cases, in addition to being constantly barraged with questions and discussions about various enhancements, bugs and other challenges.

Edward Hervey has in addition to help out with GNonlin also been working on improvements in our DVB support and improving encodebin so that you can now request a named profile when requesting pads, the last item being a crucial piece in terms of allowing me to proceed with Transmageddons multistream support.

Stefan Sauer spent time on fixing various bugs in the GStreamer 1.0 port of Buzztard and a first stab at designing a tracing framework for GStreamer.

Arun Raghavan was working on various bugs related to Pulse Audio and GStreamer and also implemented a SBC RTP depayloader element for GStreamer.

Tim-Philipp Müller has been working on implementing a stream selection flag in order for GStreamer player to be able to follow any in-file hints about which streams to default to or to not default to for that matter.

As for myself I been mostly working on Transmageddon trying to get the multistream and DVD support working. Thanks to some crucial bugfixes from Edward Hervey and Wim Taymans I was able to make good progress and I have ripped my first DVD with Transmageddon now. There is still a lot of work that needs doing, both in terms of presentation, features and general robustness, but I am very pleased by the progress made.

transmageddon1
Title selection screen, needs a bit more polish, but getting there.

Transmageddon screenshot ripping a DVD
As you see above you can now choose to transcode to different codecs for each sound stream, or drop the streams you don’t care about. The main usecase for different codecs is to you a different codecs for surround sound as opposed to stereo or mono streams.

A big thank you to Collabora and Fluendo for sponsoring us with dinner during the hackfest.

Also a big thank you to Collabora, Fluendo, Google, Igalia, Red Hat and Spotify for letting their employees attend the hackfest.

Further progress on multistream Transmageddon

As mentioned in my previous blog entry I am working on multistream handling in Transmageddon. Not been a lot of changes, but I have been able to put in a little time here and there. The changes needed to accommodate this have also cleaned up the codebase quite a bit in my opinion, moving from a forest of variables to a list of python dictionaries. This change makes keeping track of whats happening in the codepath a lot easier as I can now just print the dictionary from the list to see what all relevant values are at a given point. Anyway a little screenshot below to show where I am at:

In-progress support for multiple audio streams.

Still quite a bit of work to do to clean up the codebase and decide how certain things are to be handled (or not handled), but it is getting there. Screenshot above actually demonstrates one thing I haven’t decided on yet, which is how to deal with combining a device preset with a multistream file.

The biggest blocker currently for finishing this work is that the GStreamer encodebin element does not have an API yet for dealing with selecting encoding settings for multiple streams as detailed in this bug report. If anyone got the inclination to cook up a patch for encodebin which adds support for this that would be much appreciated.

Anyway, once I have this completed I think my next step will be to try to add some kind of DVD ripping support to Transmageddon and some basic metadata checking/editing and move the video flipping support into a special menu and add support for enabling/disabling deinterlacing in that same special menu. I trying to figure out as I go along how I can keep the user interface simple and straightforward and add requested features. The question that I continuously ask myself is what features do belong in Transmageddon and what features are of a level where people should go to something like PiTiVi instead.

Improved handling of files with multiple tracks in GStreamer

Thanks to Sebastian Dröge there is a new thing in GStreamer called streamid. It basically gives all streams inside a given file a unique id, making files with multiple streams a lot easier to deal with. This streamid is also supported by the GStreamer discoverer object. So once you identified the contents of a file with discoverer you can be sure to grab the exact stream you want coming out of (uri)decodebin by checking the pad for the streamid. The most common usecase for this is of course files with multiple audio streams in different languages.

From the output of Discoverer the stream id is really easy to get:
On the stream object you get out of Discoverer you just run a:

stream.get_stream_id()

On the pad you get from decodebin or uridecodebin the patch is a bit more convoluted, but not
to hard once you know how (there might be some kind of convenience API added for this at some point).

Before you connect the pad you get from the bin you attach a pad to it like this:

src_pad.add_probe(Gst.PadProbeType.EVENT_DOWNSTREAM, self.padprobe, None)


Then you in the function you define you can extract the stream_id with the parse_stream_start call as seen below:

def padprobe(self, pad, probeinfo, userdata):
       event = probeinfo.get_event()
       eventtype=event.type
       if eventtype==Gst.EventType.STREAM_START:
           streamid = event.parse_stream_start() 
       return Gst.PadProbeReturn.OK

I been using this code in my local copy of Transmageddon to start implementing support for files with multiple audio streams (also supporting multiple video streams would be easy, but I am not sure how useful it would be). Got a screenshot of my current development snapshot below, but I am still trying to figure out what would be a nice way to present it. The current setup will look quite crap if the incoming file got more than a few audio streams. Suggestions welcome :)

Transmageddon multistream  devshot

Transmageddon multistream development snapshot