Just been playing with my Pi2 (which has MPEG2 and VC-1 licences) and TV Headend.
Watching the Freeview SD MPEG2 576/50i broadcasts (that are natively 50i) deinterlaced scaled to 1080/50p, the ticker de-interlaces oddly with MMAL Advanced. It combs partially, with jittering dots/lines (in the 576i domain) on the edges of motion. really obvious on the BBC News channel ticker at the bottom of screen. De-interlacing is taking place, and you do get 50Hz (not 25Hz) motion, but it isn't clean. It almost looks like a de-interlacing fault.
Switching to MMAL Bob you get clean 50Hz motion with none of the odd artefacts but it is noticably softer.
It could be that MMAL Advanced just doesn't quite cut it but I thought it was better than this. Elsewhere people have suggested it is as good as YADIF 2x, and delivered good results for SD content.
(I'm not using the Half rate options, and am watching 50Hz at 50Hz so no motion judder due to mismatched frame rates.)
I've also noticed that OMXPlayer does a MUCH better job than just having MMAL enabled in Playback Acceleration with TV Headend streams?
I'm running OE 5.0.3 (OE release) - anyone got any thoughts?
(2015-02-17, 16:11)noggin Wrote: [ -> ]Watching the Freeview SD MPEG2 576/50i broadcasts (that are natively 50i) deinterlaced scaled to 1080/50p, the ticker de-interlaces oddly with MMAL Advanced. It combs partially, with jittering dots/lines (in the 576i domain) on the edges of motion. really obvious on the BBC News channel ticker at the bottom of screen. De-interlacing is taking place, and you do get 50Hz (not 25Hz) motion, but it isn't clean. It almost looks like a de-interlacing fault.
Do you have a good sample file that shows the issue.
I have a suspicion the the advanced algorithm may get the tff/bff odd/even wrong with some files.
I really need to step through the code with the debugger, but having a file that is *clearly* wrong, rather than subtly wrong would make this a lot easier.
(2015-02-17, 17:12)popcornmix Wrote: [ -> ] (2015-02-17, 16:11)noggin Wrote: [ -> ]Watching the Freeview SD MPEG2 576/50i broadcasts (that are natively 50i) deinterlaced scaled to 1080/50p, the ticker de-interlaces oddly with MMAL Advanced. It combs partially, with jittering dots/lines (in the 576i domain) on the edges of motion. really obvious on the BBC News channel ticker at the bottom of screen. De-interlacing is taking place, and you do get 50Hz (not 25Hz) motion, but it isn't clean. It almost looks like a de-interlacing fault.
Do you have a good sample file that shows the issue.
I have a suspicion the the advanced algorithm may get the tff/bff odd/even wrong with some files.
I really need to step through the code with the debugger, but having a file that is *clearly* wrong, rather than subtly wrong would make this a lot easier.
Will record some BBC News channel SD from Freeview. The ticker really shows the artefacts. It isn't presenting fields out of order (so isn't rendering tff as bff or vice versa, as the motion is correct) globally, but it could be doing it on a pixel/line-based basis.
Will try to record a short clip tonight and upload.
Popcornmix - you have mail. Have PMed you links to some short off-air BBC News channel clips with tickers on them that clearly show the difference between Bob and Advanced MMAL de-interlacing. Let me know if I can be of more help.
And thanks for all of your hard work on Kodi on the Pi platforms. Really is appreciated. I'm sure others would agree.
In progress. I've found something wrong. I believe the advanced deinterlace is using the wrong input fields in some situations, but it's a real nightmare to debug.
I'll get there though...
I wouldn’t know where to start but luckily the Pi Kodi community has you popcorn.
Its still hard to believe how great Kodi runs on the Pi when not so long a go you needed to have a big expensive and noisy PC to run it even half as well.
Just want to say thank you for all the hardwork you constantly put in.
Ditto. The whole Raspberry Pi, Kodi and OpenElec (and other distro) teams deserve a huge vote of thanks from us. It is simply amazing that a £25 credit-card sized computer is doing this, and doing it so well.
popcornmix - thanks so much for your continued hard work. If I can do anything to help - please let me know.
I've just pushed a firmware update to fix an issue with advanced deinterlace. Looks pretty close to yadif x2 with the sample files now.
You can update firmware (start.elf/fixup.dat) manually, or test with next Milhouse build.
(2015-02-21, 17:41)popcornmix Wrote: [ -> ]I've just pushed a firmware update to fix an issue with advanced deinterlace. Looks pretty close to yadif x2 with the sample files now.
You can update firmware (start.elf/fixup.dat) manually, or test with next Milhouse build.
Wow - that was quick.
How would I update a current standard OE build ?
(Would updating a Raspbian install and then copying the files work?)
(2015-02-21, 17:57)popcornmix Wrote: [ -> ]Try this:
http://openelec.tv/forum/124-raspberry-p...=15#131966
Hmm. Just tried them in OE5.0.3 by adding the start and fixup files from github.
The MMAL Advanced de-interlace no longer has the semi-random artefacts on the BBC News Channel ticker, however the edges now consistently show combing artefacts on motion on the trailling/leading edges but also look a bit blocky and field-doubled? It looks like there is aliasing on vertical motion - but not sure if this is de-interlacing or 576i->1080p scaling artefacts or both. There is obvious 50Hz motion though - so de-interlacing is happening.
The MMAL Bob de-interlace still looks better on the ticker, and I'm not convinced it looks softer. If anything the ticker looks sharper because of the lack of combing artefacts on the leading/trailling edges of linear motion? It doesn't look as aliasy on vertical motion either?
(2015-02-21, 18:23)noggin Wrote: [ -> ]The MMAL Bob de-interlace still looks better on the ticker, and I'm not convinced it looks softer. If anything the ticker looks sharper because of the lack of combing artefacts on the leading/trailling edges of linear motion? It doesn't look as aliasy on vertical motion either?
For stationary parts of video weave is the optimal deinterlacer.
For fast moving parts of video bob is the optimal deinterlacer.
yadif tries to be a compromise between the two, but I think the ticker will always look cleaner with bob.
Can you try deinterlacing with kodi (or VLC) on PC which should use yadif, and perhaps grab screenshots where PC version looks better?
To me, the ticker looks pretty similar between Pi and PC.
I have spotted a bug. bff frames are being temperally swapped (juddery). Just testing a fix.
But that won't help the BBC clips which are tff.
(2015-02-21, 18:50)popcornmix Wrote: [ -> ] (2015-02-21, 18:23)noggin Wrote: [ -> ]The MMAL Bob de-interlace still looks better on the ticker, and I'm not convinced it looks softer. If anything the ticker looks sharper because of the lack of combing artefacts on the leading/trailling edges of linear motion? It doesn't look as aliasy on vertical motion either?
For stationary parts of video weave is the optimal deinterlacer.
For fast moving parts of video bob is the optimal deinterlacer.
yadif tries to be a compromise between the two, but I think the ticker will always look cleaner with bob.
Can you try deinterlacing with kodi (or VLC) on PC which should use yadif, and perhaps grab screenshots where PC version looks better?
To me, the ticker looks pretty similar between Pi and PC.
Yep - will do. EDIT - Have PMed Kodi screenshots of the BBC News channel deinterlaced on a Pi and a Chromebox with both MMALs on the Pi, YADIF and VAAPI MCDI on the Chromebox.