View Full Version : Importing DNxHD to Avisynth
SAPikachu
2nd July 2012, 12:12
Not sure whether the official indexer is compatible, it is better to use the one in my build, just uploaded (http://nmm.me/l9).
kolak
2nd July 2012, 14:29
I'm not sure neither- I think it's not compatible.
Thanks- looks like indexing over network is very slow, even if I have proper gigabit network which works fine on Win 7.
kolak
16th July 2012, 12:00
Here is some sample so you can play with it (comes from the same file as text samples)
http://www.sendspace.com/file/v86w3w
Look at the artefacts around text.
Any news about ffms2 DNxHD10bit decoding bug (or whatever is causing it)?
Keiyakusha
16th July 2012, 18:22
Sadly no. There is only 2-3 ways:
1) get ffmbc dev make patch for upstream, but he is not willing to do so
2) get ffms2 devs to fix it again, but this is not easy too.
3) make ffmpeg devs fix it, impossible. will take ages to commit even if it will happen.
kolak
16th July 2012, 20:19
:( Can't we use ffmbc as an engine for ffms2 instead of ffmpeg?
I thought every change to ffmpeg code is publicly available- no?
Keiyakusha
16th July 2012, 20:40
I thought every change to ffmpeg code is publicly available- no?
It is (well at least it should be), but it is a huge thing. As far as I understand there is no easy way to figure out what and where was changed to fix the issue we interested in. it shouldn't be hard for ffmbc dev to point out where is the relevant code though. Maybe try to ask him too.
poisondeathray
16th July 2012, 20:45
:( Can't we use ffmbc as an engine for ffms2 instead of ffmpeg?
I thought every change to ffmpeg code is publicly available- no?
Another good reason for this - ffmbc handles many interlaced h.264 transport streams much better than ffmpeg for whatever reason
In the ffms2 thread there are many complaints about this known issue
SAPikachu
17th July 2012, 04:10
:( Can't we use ffmbc as an engine for ffms2 instead of ffmpeg?
I thought every change to ffmpeg code is publicly available- no?
That's a good idea, how can I miss this...
Just compiled a build against latest ffmbc (0.7rc7), seems it decodes without problem but I didn't check whether the bug is fixed. Anyone interested to try it?
(Deprecated, please see http://forum.doom9.org/showthread.php?p=1583307#post1583307 for new version)
Keiyakusha
17th July 2012, 07:56
That's a good idea, how can I miss this...
Just compiled a build against latest ffmbc (0.7rc7) (http://nmm.me/mc), seems it decodes without problem but I didn't check whether the bug is fixed. Anyone interested to try it?
Yep, looks like it was. Thanks!
Old build (http://i.imgur.com/ZysFo.png) vs. New build (http://i.imgur.com/FWtLR.png) (make sure browser doesn't performs resizing of the image or it won't be seen.)
However I'd say LSB looks worse now.
SAPikachu
17th July 2012, 08:14
Yep, looks like it was. Thanks!
Old build (http://i.imgur.com/ZysFo.png) vs. New build (http://i.imgur.com/FWtLR.png) (make sure browser doesn't performs resizing of the image or it won't be seen.)
However I'd say LSB looks worse now.
I'm not sure, but it looks like encoding artifacts?
Keiyakusha
17th July 2012, 10:34
I'm not sure, but it looks like encoding artifacts?
Hmm, I don't think they are artifacts. The video source is a still image under 185mbps bitrate... it should be virtually lossless. Image taken with old build have straight lines which Is expected. Bot on another screenshow... I never seen such artifacts, pattern looks too perfect to me and doesn't even reminds any kind of macroblocks.
It will be nice to hear someone else's opinion on this.
kolak
17th July 2012, 12:07
That's a good idea, how can I miss this...
Just compiled a build against latest ffmbc (0.7rc7) (http://nmm.me/mc), seems it decodes without problem but I didn't check whether the bug is fixed. Anyone interested to try it?
I will check it.
update: it works fine- great work :)
:thanks:
kolak
17th July 2012, 13:23
So now- does ffms2 has to index all files, even those with I frames only- like ProRes, uncompressed?
Indexing is bit slow- I have 0.5% CPU usage.
Some QT reference files are also not working same way as in QT player.
SAPikachu
18th July 2012, 01:28
So now- does ffms2 has to index all files, even those with I frames only- like ProRes, uncompressed?
Yes, ffms2 need index for all files. Indexing speed mainly depends on your disk speed (or network speed if your file is on a network drive).
Indexing is bit slow- I have 0.5% CPU usage.
You can check transfer speed using performance monitor (assuming you are using win7). If indexing uses more than half available bandwidth then I think it is normal.
Some QT reference files are also not working same way as in QT player.
How are they different?
PS: The previous version can't read color matrix flag from source file, I just compiled a build that fixed that (http://nmm.me/me).
kolak
18th July 2012, 10:22
Sorry- indexing is fine, I'm just not patient enough :)
It reads at about 100MB/sec. Gigabit on Win7 works at full speed- not at 50% like it use to be on XP.
One thing- in some cases info jumps from 0 to 99% straight away, but sometimes it works fine. These are all I frame only files- ProRes, DNxHD, uncompressed.
It took about 6 minutes to do 130GB ProRes on my local RAID, so it's ok.
Files, which I refer to are reference files to AVID MXF files (reference files are created by 3rd party software CalibratedQ). I noticed that nothing based on ffmpeg can open them- ffmpeg, ffmbc etc. They do work fine in QT player.
Thanks for the new build- reading ProRes, DNxHD was fine anyway (at least for my samples), but good to have it working as it should.
jmac698
1st August 2012, 19:13
Cool, updated http://avisynth.org/mediawiki/High_bit-depth_Support_with_Avisynth
SAPikachu
2nd August 2012, 01:13
Cool, updated http://avisynth.org/mediawiki/High_bit-depth_Support_with_Avisynth
Thanks. :) But I think we need to mention that ffms in the first post is linked with libav, and the second one is linked with ffmbc. The first version may be better for general video files, but have bug on decoding DNxHD files. The second one fixed the bug, and may have better support for professional video formats.
jmac698
2nd August 2012, 04:07
Um... could you go there, click edit, and make the changes? It's not hard :)
SAPikachu
2nd August 2012, 07:21
Um... could you go there, click edit, and make the changes? It's not hard :)
Oh sorry, previously I thought that it needs special privilege to edit the wiki. Just made some modification to it. :)
jmac698
2nd August 2012, 07:59
Thanks.
I made a small correction, but really I should have told you first and let you do it. You can check if you agree.
SAPikachu
2nd August 2012, 08:05
That's OK, it was just my typo (and bad grammar :P ). The paragraph is clearer now, thanks. :)
fleebitten
17th January 2013, 11:40
I have a problem loading the patched ffms2.dll.
I'm using Avisynth 2.60 and Virtualdub 1.10.3 (32bit). I have ffms2.dll, ffmsindex.exe and FFMS2.avsi in my Avisynth plugins folder. Opening my avs file in Virtualdub gives the following error:
Avisynth open failure:
LoadPlugin: unable to load "C:\Program Files\AviSynth 2.5\plugins\ffms2.dll", error=0x7f
However, using an unpatched ffms2.dll and ffmsindex.exe works. There is a delay while indexing then the video opens.
I've tried both versions: ffmbc and libav. I used LoadCPlugin for ffmbc which I believe is correct. I admit I lean towards the newbie side.
Any ideas?
Many thanks.
SAPikachu
18th January 2013, 12:52
I have a problem loading the patched ffms2.dll.
I'm using Avisynth 2.60 and Virtualdub 1.10.3 (32bit). I have ffms2.dll, ffmsindex.exe and FFMS2.avsi in my Avisynth plugins folder. Opening my avs file in Virtualdub gives the following error:
However, using an unpatched ffms2.dll and ffmsindex.exe works. There is a delay while indexing then the video opens.
I've tried both versions: ffmbc and libav. I used LoadCPlugin for ffmbc which I believe is correct. I admit I lean towards the newbie side.
Any ideas?
Many thanks.
Sorry I mistakenly broke XP compatibility of my last build. Are you using Windows XP? If so, you can try my latest build (http://nmm.me/vh).
fleebitten
18th January 2013, 20:30
Hello SAPikachu, yes I'm using XP. I'll give this one a try, thanks very much.
fleebitten
19th January 2013, 11:38
Just to confirm that this worked okay, thank you.
Does this version support 4:2:2 like the previous one you posted?
Many thanks.
SAPikachu
19th January 2013, 13:36
Just to confirm that this worked okay, thank you.
Does this version support 4:2:2 like the previous one you posted?
Many thanks.
Yes, hacks are the same as ones in previous versions.
fleebitten
19th January 2013, 16:43
That's great, thanks.
I would just like to check that everything is okay as I'm not sure exactly what the LSB part should look like. I'm opening a 10bit 4:2:2 v210 clip. Viewing in Virtualdub shows green for the LSB part with just the brightest highlights that are in the video coming through (white only). In other words, just the very highest luma values. Is this how it should be?
SAPikachu
20th January 2013, 12:02
That's great, thanks.
I would just like to check that everything is okay as I'm not sure exactly what the LSB part should look like. I'm opening a 10bit 4:2:2 v210 clip. Viewing in Virtualdub shows green for the LSB part with just the brightest highlights that are in the video coming through (white only). In other words, just the very highest luma values. Is this how it should be?
For 10bit videos there aren't much data in LSB part, so as long as LSB part is not all green, it should be normal.
fleebitten
21st January 2013, 22:40
Thanks again SAPikachu.
I've a problem with f3kdb_dither though. My source video is v210 4:2:2 10bit as mentioned. With the following code:
FFVideoSource("test.avi", enable10bithack = true)
f3kdb_dither(input_depth = 10)
I get a lot of noise and no picture to speak of. Not until I set input_depth to 16 do I get a normal picture.
Any ideas why that is?
Thanks for any help.
kolak
21st January 2013, 23:06
Maybe because hacked ffms2 reads 10bit as 8bit+8bit, so input should be 16bit. Avisynth does not support "real" 10bit formats.
SAPikachu
22nd January 2013, 01:24
Yes, enable10bithack will automatically convert source to 16bit for easier previewing, so you should specify input_depth=16 in f3kdb/f3kdb_dither.
leoenc
22nd September 2013, 14:57
With enable10bithack I'm getting a double height frame (1920x2160). I tried both progressive and interlaced sources, and AVC-Intra and DnxHD sources. I tried ffms2-9417b9c-libav-aca2510 and ffms2-r734-libav-e6160bd.
The bottom 1080 part is filled with colorful blocks and cropping resolves the issue.
Is that expected behavior?
cretindesalpes
22nd September 2013, 19:52
Yes, it’s the “stack16” 16-bit format. Use DitherPost() if you want to go back to 8 bits, or don’t set enable10bithack.
leoenc
23rd September 2013, 07:00
Ok, thanks
MysteryX
23rd December 2015, 04:32
It seems this version of FFM2 doesn't support VP9 videos, while the official version does.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.