methods for opening MP4 and MKV in avisynth

[TheFluff]

Originally Posted by neuron2

False. If the timestamps are not fully correct and continuous it fails. It also fails for elementary streams.

mkv's and mp4's should always have correct timestamps, if not it is a corrupt file and all bets are off regardless of source filter

and this thread is not about elementary streams now is it

[Guest]

I responded to your statement, which did not contain any qualifiers, and which explicitly referred to other formats.

[poisondeathray]

Maybe you can add that another potential pitfall of DSS/DSS2 is that it predisposes you to all the filters in your directshow chain. e.g. if you forget to turn off processing in ffdshow (you might have used for playback) , etc... those effects get passed through when you encode. So if you are not very diligent you can get burned.

Also when using DSS/DSS2, the choice of decoder becomes more important, and each may behave differently or have quirks. e.g. there are differences between using ffdshow vs ffdshow-mt vs. coreavc vs. divx h.264 etc...

[TheFluff]

Originally Posted by neuron2

I responded to your statement, which did not contain any qualifiers, and which explicitly referred to other formats.

hmm, yes, a semantic argument, how productive

Does DSS2 even open elementary streams at all? What directshow source filter could be used to open them? I already stated that DSS2's capabilities are obviously dependent on your system's dshow splitters/decoders.

I obviously didn't qualify "this filter may not work correctly with broken files herp derp" because I thought it was self-evident, but apparently not!

[halsboss]

Originally Posted by TheFluff

you got about half of them wrong

Well, help fix them with some facts.

Originally Posted by TheFluff

and turned this into a neuron2 advertising thread, congratulations

my, my, a little pointed aren't we. That isn't my intent, why would you want to suggest so ? Don't answer that.

Originally Posted by TheFluff

it's funny how you mention the open gop issue that can happen with m2ts as if it was relevant for mkv/mp4;

No it isn't funny, you SAW I am asking for clarification. It isn't relevant for mkv/mp4 ?

Originally Posted by TheFluff

it's also funny how you think you can determine the truth only by reading doom9 posts

It's a good a means as any, reading contributions of people who know things, including you. Why else have a forum. Why are you so emotionally involved ? If you think you can fix it up then please by all means clarify. Some info appears to conflict with others and that's been raised above too, for knowledgeable people like you to collaborate and clarify for the betterment of others who use the forum.

PS I see I omitted a googled link I came across re the possibility DSS2 and "glitchy" (not non-frame-accurate) frames after a seek, will see if I can find it again. Nope, lost.

PPS please post a better table if you like. Just food for thought, should, probably good enough aren't entirely comforting. that may be life, but does it change the table ? If so please feel free to change it.

[TheFluff]

Originally Posted by halsboss

No it isn't funny, you SAW I am asking for clarification. It isn't relevant for mkv/mp4 ?

Not really, no.

Originally Posted by halsboss

It's a good a means as any, reading contributions of people who know things, including you. Why else have a forum. Why are you so emotionally involved ? If you think you can fix it up then please by all means clarify. Some info appears to conflict with others and that's been raised above too, for knowledgeable people like you to collaborate and clarify for the betterment of others who use the forum.

Unlike peer reviewed scientific papers, doom9 posts aren't really all that reliable (just the likelihood of them being outdated is a significant factor) and secondary-source research based on them cannot be considered as relevant to reality. If you want to find something out, test it yourself.

[halsboss]

Originally Posted by TheFluff

Unlike peer reviewed scientific papers, doom9 posts aren't really all that reliable (just the likelihood of them being outdated is a significant factor) and secondary-source research based on them cannot be considered as relevant to reality. If you want to find something out, test it yourself.

Good thoughts. Mostly people pragmatically go with what others, whom they choose to consider knowledgeable, say rather than reinvent the wheel; and this stuff isn't going to end the world. The objective of the thread was to tap into others knowledge as it is today, raising and putting to rest furfies and obsolete info.

So, DSS2 is frame accurate, then and it relies on a well behaved and understood splitter and filter chain (haali, coreavc?). And seeking to the middle of a GOP (?) won't return dud looking frames because it's not relevant to mkv/mp4. And ffmpegsource is similar (albeit the words above re libav are very much less than confidence-building). There was a little recent discussion here http://forum.doom9.org/showthread.php?t=139092 but I'm not sure of a conclusion.

[poisondeathray]

I use DSS2 frequently, and it works fine. Another quirk is that the fps is often not perfect, but off by a bit. (it's advisable to check with info() in avisynth), and one method is to use AssumeFPS() to correct it. e.g AssumeFPS(24000,1001) for film

[TheFluff]

Originally Posted by halsboss

So, DSS2 is frame accurate, then and it relies on a well behaved and understood splitter and filter chain (haali, coreavc?). And seeking to the middle of a GOP (?) won't return dud looking frames because it's not relevant to mkv/mp4. And ffmpegsource is similar (albeit the words above re libav are very much less than confidence-building). There was a little recent discussion here http://forum.doom9.org/showthread.php?t=139092 but I'm not sure of a conclusion.

I reiterate:

Originally Posted by TheFluff

ffmpegsource has never used libavformat to open mkv's

It has always used Haali's matroska parser, which doesn't have any particular random access issues. I have never seen any random access issues of note with mp4's and lavf either; what Myrsloik mentioned in the ffms2 thread mostly applies to less well supported/specified containers.

As for the open GOP issue, the entire thread seems focused on h264, and h264 doesn't even have open GOP's as such, AFAIK (IDR-frames and multiple references kind of ruin the idea). You're also unlikely to ever run into an MPEG4 ASP stream with open GOP in the wild (you weren't thinking of downloading stuff, were you, you naughty creature?) unless you encode one that way yourself (so don't do that). I wouldn't ever put MPEG2 into MP4 or MKV either so that is out too. In either case, if you do somehow run into an open GOP with DSS2 the splitter (if it is a good splitter) is likely to back up to nearest reliable decoding start point for you anyway. I imagine the reason it doesn't do this with M2TS is that Haali's TS splitter isn't all that perfect (yet).

[halsboss]

Excellent info, thankyou people. Will post an updated summary soon, pending any more clarifications.

edit: points are well made, although I admit phrases like "(if it is a good splitter) is likely to" don't inspire confidence it'll actually do the right thing; albeit no reports of bad things it seems.

[vie2233hil]

What's the difference between these options quality-wise? directshow ones could use coreavc which is thougth of having less quality output than libavcodec based options? Does libavcodec provide good quality?

Originally Posted by vie2233hil

What's the difference between these options quality-wise? directshow ones could use coreavc which is thougth of having less quality output than libavcodec based options? Does libavcodec provide good quality?

Quality doesn't apply to AVC decoding, every decoder is based on the same specifications, and therefor, barring any bugs, should have identical output ...

Just to further the discussion also, DirectShowSource(), the original one, is pretty bad for HD (720p/1080p) MKVs, the last time I tried it (even with Haali/CoreAVC, Haali/FFmpeg and Haali/QTdecode) as a "straight through" encode, there was the occasional (like one every 10-15secs) frame jump / misplacement ... at the time I rectified the problem with the original FFmpegsource() ... haven't completely tested version 2 of either ...

Since I don't own a Nvidia 88xx series or better card, am waiting to see how this thread concludes

7ek

[Mr VacBob]

Originally Posted by TheFluff

As for the open GOP issue, the entire thread seems focused on h264, and h264 doesn't even have open GOP's as such, AFAIK (IDR-frames and multiple references kind of ruin the idea).

h264 can emulate this pretty well; it has recovery point SEIs, which streams can use to provide seek points without IDR. I can't remember if ffmpeg supports them (or if it's even conceptually possible to support them).

And of course, if there's some PAFF stream lavc doesn't decode, or some problem preventing DGAVCDec from upgrading lavc, I hope it's in ffmpeg roundup...

[medp7060]

Originally Posted by halsboss

Excellent info, thankyou people. Will post an updated summary soon, pending any more clarifications.

edit: points are well made, although I admit phrases like "(if it is a good splitter) is likely to" don't inspire confidence it'll actually do the right thing; albeit no reports of bad things it seems.

have you updated the summary?