You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

162 lines
6.3 KiB

ScreenCast with transcode and import_x11
======================================================================
Francesco Romani <fromani at gmail dot com> 18 August 2006
*********************************************************************
*** WARNING: ***
*** import_x11 module is still EXPERIMENTAL! ***
*** It may DON'T work for you or it can deliver BROKEN pictures! ***
*** If you experiences strange (= not explained here) behaviour, ***
*** PLEASE drop a note on transcode-devel@exit1.org explaining ***
*** your issue. Thanks! ***
*********************************************************************
1. What is this?
----------------
import_x11 is the new (as in transcode 1.1.0 and later) module and
infrastructure designed to allow to capture full-screen images from
a running X-server, in order to partially replace the aging and
not-widely-appreciated import_vnc module.
import_x11 isn't a drop-in replacement for import_vnc since vnc
runs nicely on more platforms that X11 (i.e.: windows).
Despite this, import_x11 features a much better support and (at
least in intentions) a cleaner design and implementation.
Please also consider the fact that transcode itself DOES NOT
run on windows/non-posixly platforms, so X11 import module it's
supposed to cover most of use cases of import_vnc.
2. Basic usage
--------------
transcode's X11 sourcing facilities are accessible via transcode's
main program or via tcprobe support utility.
tcprobe allows to see the properties of any LOCAL X11 connection.
That is usefull mailny for transcode internal usage to figure out
(pseudo)stream properties, but can be also useful for debug purposes.
usage is straightforward, just supply the (LOCAL) DISPLAY identifier
as input argument, like following example
tcprobe -i :0.0
that produces, on my box (as for transcode 1.1.0):
summary for :0.0, <*> = not default, 0 = not detected
stream type: X11 display source
video format: rgb
import frame size: 1024x768 [720x576] (-g) <*>
aspect ratio: 1:1 asr=1 (--import_asr)
frame rate: 10.000 [25.000] frc=11 (-f) <*>
no audio track: (use "null" import module for audio)
Please note that import frame rate it's just a kind suggestion and
can be overloaded with no theroretical penalty except for performance
ones.
In order to use X11 source with transcode, a little more is needed.
Input source must be specified in the same format as for tcprobe,
and internal colorspace must be set as RGB24, using (as for transcode
1.1.0) -V rgb24.
example:
transcode -i :0.0 -V rgb24 [other options]
3. Recording a session
----------------------
As outlined in example above, recording a screen cast using transcode
is pretty straightforwarded, since it's matter of selecting a proper
X11 source and change internal image representation to rgb24.
Anyway, there are some issues that must be understanded to avoid
unpleasent surprises.
As well as much of multimedia processing, there are two main concerns:
processing speed and storage size.
First of all, take in mind that X11 connections delivers often BIG
(1024x768 or even more larger) pictures that aren't trivial to elaborate
(of course in small amount of times) in plain RGB24 format.
But since most of popular codecs out there works on YUV flavours,
a format conversion is needed. Such conversion take time and CAN
lead to substantial frame rate drop. It's definitively possible
that transcode just can't capture images at rate superior to a certain
rate on your box.
That's partially due to some internal transcode, that are under
investigation and/or under improvement, but that's also caused by X11
structure and by the nature of the pictures involved on the process.
On the other hand, NOT compressing the capture stream quickly produces
HUGE files. Example, with a display running at 1024x768:
one picture (one frame): 1x1024x768x3 = ~2304 KBytes (~2.25 MBytes)
TWO seconds at PAL fps recording (50 frames total):
~115200 KBytes (~112.5 MBytes)
TEN seconds at PAL fps recording (250 frames total):
~576000 KBytes (~562.5 MBytes)
and so on.
Of course encoding in raw format delivers a very good encoding speed (read:
introduce a very few overhead in process except for bandwith usage), but
it isn't an handy option for space reasons described above.
Using a lightweight encoder like LZO (see documentation of lzo modules)
can be usefull.
The other problem is recording rate. There a some factors that severely
limit the acquisition rato (so the encoded clip fps rate), notably
the processing power required for each frame, and that is function of
1) frame conversion (EX: RGB24 -> YUV420P) performed
2) filter applied, if any (EX: antialiasing)
3) encoder used (EX: xvid vs lzo)
The other factor is the efficiency of transcode's X11 bridge module.
The latter factor can be significantly improved, the former can hardly be.
So, if you want/need high-fps screen recording and transcode seems unable to
achieve this task, try first to perform a two-pass recording before reporting
the problem on ML :)
first pass: just grab the screencast doing the fewest processing possible,
and using the most lightweight encoder avalaible.
second pass: apply all further processing that you want, offline.
4. Known issues
---------------
Only 24-bit depth running X servers are supported. That's a pretty
severe and annoying constraint and it will go away in future releases.
Only LOCAL (not-networked) X connections are supported. That's was
a design choice and isn't likely it will be changed soon.
Frame timing could be inaccurate or just wrong. We're working on this,
but isn't trivial to solve. Stay tuned.
Communications with X server is pretty inefficient, yet, since it doesn't
use any of advanced X features like Shared Memory, extensions and so on.
So, communication consume a fair amount of bandwith.
This can lead to issue above and in general to a slowdown of machine.
This will be solved or at least improved in future releases.
5. Future plans
---------------
import_vnc will not go away soon. It will stay at least for 1.1.0
release cycle and most likely for 1.2.x/post-1.1.x too.
Neverthless, it's legacy and very poorly mantained (and maintanable)
code, and it will be blasted away in the long shot.