FFmpegSource

[LoRd_MuldeR]

Here is the sample:
http://www.mediafire.com/file/4wuhzj...ample.ffms2.7z

[Myrsloik]

This was an interesting clip. What internally happens:
1. Seeking is required
2. The normal LAVF seeking method is called
3. A packet is attempted to be read but fails since 2 failed and the position is now at the file end?
4. No audio for you

For some reason it doesn't like your audio track. Adding AVSEEK_FLAG_ANY seems to make the seeking succeed on your clip but seems to introduce audible artifacts in my other mp3 in avi test clip. I think this is a LAVF bug so remux to mkv and enjoy life.

[Myrsloik]

Some more thinking produced this compile. It should work on more clips like yours but without introducing any regressions at least. So much for seeking always succeeding in a simple avi...

[LoRd_MuldeR]

Thanks. I will test your new compile once I get home. I mist to leave now.

BTW: That AVI file was created with Avidemux.

[Myrsloik]

This is the full todo list for FFMS2. Vote on the features you want or suggest your own. I'm personally guessing 2-5 are what people want mostly but I could be wrong...

Todo

1. (Implemented) Make it possible to set the output colorspace easily (could be useful for files decoded to full resolution YUV or another format not directly supported by avisynth, implementation-wise related to 8)
2. (Implemented) Properly detect nvops so many normal xvid files can be decoded properly (I have no idea how, really)
3. (Implemented) Detect when the file the index is for has changed (will throw an error instead of being undefined for now)
4. Write a graphic indexing app using QT creator (it will look at least twice as good as DGIndex)
5. (Implemented) Show progress in haalisource indexing (assuming Haali reveals the secrets)
6. (Implemented) Make stream delays optional to include in the filenames from avisynth/have configurable filenames in some way
7. Support uncompressed formats and formats that need codec properties to be set from the container values properly (added width and height reading for all formats, currently only supported in LAVF opened files but wav in mkv/ts is so uncommon I don't think anyone's noticed)
8. (Implemented) Handle video with resolution changes (now I have samples, closely related to 1)
9. (Implemented) RFF flag support - not compatible with VFR but that's not my problem
10. Implement incremental indexing (this would be a pure API improvement and never show up in avisynth because of its limitations)
11. (Implemented) Use unicode for all opened and saved filenames
12. (Added API documentation/revised the avisynth manual) More documentation
13. (Implemented) Add VS2008 project files and cmake for all other systems to use
14. Regression tests
15. (Not my fault) Figure out why Haali's parser just stops on some huge transport streams (This is a bug in Haali's code and will be fixed in a later release of his splitter)
16. Add support for transport streams that do not start on a packet boundary with Haali's parser (I probably won't implement this but patches are welcome, it could still be useful to reduce the number of needed steps to open some files)
17. Make wrappers for using the FFMS2 library in other applications such as virtualdub and premiere import plugins if there's any real interest
18. Make all calls to SWScale and PP multithreaded
19. Export metadata information in the API (needed for 4 to be more user friendly)
20. Figure out why h.264 in TS/PS decoding shows artifacts
21. Implement raw mpeg 1/2/4 and h.264 parsers to support those input formats for more than linear reading
22. Dummy d2v file writing for use with rffmode=1+TFM
23. (Implemented) Improve the horribly inefficient index format, currently a ~5MB index becomes 20kb after zipping

[TheFluff]

2, 5, don't care about the rest

[Chevy787]

5,4, 12 (for added functions).
Some very nice ideas there Myrs.

[jmartinr]

2, 6, The automation would make it just a little bit easier to use.

[LoRd_MuldeR]

Originally Posted by Myrsloik

Some more thinking produced this compile. It should work on more clips like yours but without introducing any regressions at least. So much for seeking always succeeding in a simple avi...

Yes, works fine

[TheRyuu]

See here...

Updated to latest ffmpeg-mt and ffms2 (beta 9) sources as of today.

Built with ICL 11 and includes static pthreads and libfaad.

[Mr VacBob]

You can use AVCodecParser (see mpegvideo_parser.c) to discover field flags instead of having to do a complete decode.

[Myrsloik]

Now give me a code sample too because I've looked at it before but couldn't figure out what the arguments are even supposed to be or if I created it correctly. Otherwise I'm just going to assume it's something old and forgotten in FFmpeg...

[Mr VacBob]

av_parser_init(CODEC_ID_MPEG2VIDEO) -> set s->flags to PARSER_FLAG_COMPLETE_FRAMES -> av_parser_parse2 -> read s->repeat_pict?

I don't know what happens if you don't pass in real values for pts/dts/pos, though.
And why isn't av_parser_init documented, or called av_parser_open...

[vlada]

My vote: 2, 5, 4, 12

[Adub]

Originally Posted by TheRyuu

ffms2-mt-06022009.rar

Updated to latest ffmpeg-mt and ffms2 (beta 9) sources as of today.

Built with ICL 11 and includes static pthreads and libfaad.

Yes! Thank you so much for this! I was just looking for it when I came across this last page. Thanks for building it!

Edit: Okay, time for a bug report. When loading a Blu-ray M2TS file, 19.6 GB (21,077,882,880 bytes), I receive the following error upon completion of the indexing stage.



This is with the latest, ffmpeg-mt build by TheRyuu.

Video is H.264, 23.976 fps.

Script:

#DSS2("00000.m2ts")
FFVideoSource("00000.m2ts")

It seems that it works with a smaller sample as well.

Download, extract, and play the avs with the latest version of the ffmpeg-mt build.

www.mirror.adubvideo.net/ffms2bugsample.7z

[saint-francis]

Is the .m2ts playable? I can't play your sample in MPC.

[Adub]

Yeah, plays just fine in MPC HC 1127.

[Myrsloik]

I looked at the sample and can't explain it. For some reason it fails to decode even a single frame in the video which means the height/width can't be determined either. It "works" if lavf's ts splitter is used but then you have so many other issues it's not interesting anyway. Nice clip to have for future testing.

[Adub]

Damn. So no quick fixes, eh? And I was so excited to use a frame accurate ffmpeg-mt Avisynth source as well. What's weird is that it works just fine when using Haali's DSS2, and since you are using Haali's internal splitter, shouldn't it work just the same?

[Myrsloik]

Both yes and no, I only use the parser part through COM so slightly different behavior is to be expected (just like you don't see segment linking and all that fancy stuff for matroska). I think I'll have to ask Haali for ideas. Demuxing it to a raw stream makes it crash too. Is the clip corrupt?

And about your post in another thread, the mt compile doesn't have any known bugs that aren't also in the normal compile.