Guide to convert BD 3D to 3D Left+Right Stereoscopic and Anaglyph

[slavanap]

Well, the beta (i hope so) version of ssifSource lib:
http://sendfile.su/817501

Changes:
* now all ssifSource solutions is combined in this lib.
ssifSource, ssifSource2 and ssifSource3 functions are available.
* fixed the issue with decoding speed throttle.
* framecount autodetect algorithm improved.
* left_view & right_view parameters renamed to avc_view & mvc_view.
* added swap_views parameter.

About the fuctions:
The old one:

ssifSource(ssif_file, width, height, frame_count, left_track, right_track, left_264, right_264, show_params)

Parameters description are available in previous ssifSource package. This fuction uses ldecod_mod.exe & eac3to.exe (the modified one for proper pipes support).

ssifSource2(ssif_file, frame_count, avc_view, mvc_view, horizontal_stack, swap_views)
ssifSource3(filelist, avc_view, mvc_view, horizontal_stack, swap_views)

Actually ssifSource3 uses ssifSource2 to display each of the files in the filelist.

filelist - string parameter, the list of files that will be displayed sequentially. filelist format is filename1;framecount1;filename2;framecount2 and so on.
filenameX - is filename (related or fully qualified) of the ssif file. framecountX is the framecount in that file. if you specify 0 as framecount, then ssifSource2 will try to autodetect framecount of this file.

avc_view, mvc_view - boolean. Defines that base and/or dependent views should be displayed.

horizontal_stack - boolean. It makes sense only when avc_view and mvc_view are true. If horizontal_stack is true, views stacks horizontally, overwise - vertically.

swap_views - integer. It makes sense only when avc_view and mvc_view are true. Defines in which order views are displayed. If swap_views = -1 then autodetection algorithm based on mpls files tries to detect whether or not them should be swapped. If swap_views = 0 then views are not swapped, if = 1 then they are swapped.
Explanation for horizontal_stack = true:
swap_views = 0 -> base (avc) view is on the left, dependent (mvc) view is on the right
swap_views = 1 -> dependent view is on the left, base view is on the right.

Note: coreavc version 3.0.0.2 & mpegsplitter version 1.5.2.3102 is required for ssifSource2 & ssifSource3
(also works with older ones).

Please, do not rely on accurance of framecount autodetect and frame seeking. (In my testing I get the worst accurance is 4 seconds of video). I really don't know how the decoder works, I just use the data it returns in the most effective way. So what I gonna do next is adding a create-text-based-index feature in ssifSource2&3 to dump directshow timings for each frame. (actually, decoder does not support frame seeking, and my solution is based on processing of directshow timings).

I hope, that description of the new features would be enough.
If you'll find any bugs, please let me know.

ADDED:
Default values for ssifSource2&3 parameters are:

frame_count = 0 // forces autodetect
avc_view = false
mvc_view = true
horizontal_stack = false
swap_views = -1 // forces autodetect

If you did not specify any of these parameters, the default value is used.

[jdobbs]

@slavanap

I'm running the newer version on a job with a huge number of segments right now... it's working great, with an impressive speed improvement over the last release. Sweet. I'll let you know how it works out.

[Thalyn]

Originally Posted by slavanap

Can you try the new ssifSource lib with framecount autodetection on that source? It may fail, of course.

*snip*

I think I have to test it by my own, to get to know what happens with CoreAVC during processing this film. Where can I get the source you use?

The source was apparently the where my problems were. Source being an original Universal Solder: Day of Reckoning 3D BluRay.

Seems if you use the SSIF directly (including copying it to a HDD) than it gives all of the problems I had - the stall, the EAC3To errors and the bizarre framecount. Working from the two M2TS files instead gave me a good source and subsequently clean results. So I'd personally chalk that one up to just something weird with how the disc was mastered.

I'll give the new plugin a try as soon as my current transcode finishes. I got a few I need to re-do as I originally used H264StereoSource (had to trim frames), so they should give the auto-detect and seamless branching a good work-out.

[r0lZ]

Originally Posted by slavanap

About the fuctions:
The old one:

Parameters description are available in previous ssifSource package. This fuction uses ldecod_mod.exe & eac3to.exe (the modified one for proper pipes support).

Can you upload the two exes or give the link to download the previous ssifsource package? I'm not sure I have the correct versions of the exes.

[PurpleMan]

I was just wondering, is there currently any advantage in using ssifSource over DirectShowMVCSource (based on coremvc)?

(Other than the fact it costs money)

[Nico8583]

ssifSource2 uses CoreAVC dll but CoreAVC is free ?

[jdobbs]

Originally Posted by Nico8583

ssifSource2 uses CoreAVC dll but CoreAVC is free ?

No. But you can get it fairly cheaply by buying "Stereoscopic Player" at www.3dtv.at (I did). It costs about 39 euros. You used to be able to get it with "MVC to AVI" for 19 euros -- but that price has risen to 29 euros now.

[frencher]

Update my package ultimate Intel MVC Decoder x86 & x64 & SSE support + src

[Nico8583]

What does your package do or permit ?

[r0lZ]

It does MVC decoding, encoding and transcoding. I'm currently testing it. It is open source. IMO, it could be the best solution for us, as it has several advantages over the current method: open source, official updates by Intel, no need for a "magic path", no copyright problems, can probably be converted to an avisynth plugin relatively easily.

Currently, it can only decode a CombinedMVC.264 file to 2 uncompressed YUV streams, and therefore it requires MUCH disc space (for example, it requires about 800 GB to convert Ice Age 3 to YUV!) But with some changes in the code, it should be possible to output the 2 streams to pipes, and implement an avisynth plugin working like the old ldecod. Or, with a bit more work, it can certainly be implemented as a totally independent avisynth plugin. I suppose that it can also be modified to accept the 2 AVC and MVC streams produced by eac3to without any need to combine them first.

However, it seems that it has a bug with Ice Age 3. I will confirm that when I'll finish to encode it. The bug produces artefacts in the left and right borders of the image, especially in scenes with fast moving objects or fast camera movements. But it must be noted that the ssifsource2 and DirectShowMVCSource methods have exactly the same bug! (And the ssifsource1/ldecod method doesn't work at all with that BD.) I guess Peter uses the Intel SDK for his CoreAVCDecoder.

Anyway, thanks to Frencher for his great finding. (He did also the batch files in the package, useful to understand how the various commands work.)

[Nico8583]

So package is not modified ? I believed Frencher had made some changes in the source code
I've tried to decode / encode with Intel MVC but there are 2 problems :
- MVC Decoder seems to be OK (but need much disc space)
- MVC Encoder only support a space color that not corresponding with Blu ray space color -> Colors not corresponding
- MVC Transcoder gives a bad quality (details, colors) with "quality" parameter
I'm waiting for your test result

[r0lZ]

I don't think we should use it to encode. The MVC decoder only is really useful for us.

[Nico8583]

MVC Encoder/Transcoder should be used to create true BD 3D (and not SBS/TB)

[r0lZ]

Here is an example of the bug in CoreAVCDecoder (MVC stream only, of course) and probably also in the Intel SDK MVC decoder. (Still to be confirmed.)



Look at the right side of the picture. There is also a white rectangle at the left side, but usually, the artefacts are more important in the right side.
This BD has been decoded with DirectShowMVCSource, but I did also a test with ssifsource2 and a decoding from a CombinedM2TS file, just to be sure, and they have the same problem.
When decoded and encoded with DVDFab, there is no problem. (The image quality is terrible, but it's DVDFab!)

Please wait for the Intel SDK test. I need to use my computer for other things right now, and I'll launch the encoding tonight...

[Nico8583]

DVDFab is not the best software for video but is not the poorest, source code could be a good thing but it is not available...

[r0lZ]

I know. I hate DVDFab's h264 encoder, but the other things it does can be useful.
I installed and used it only to test if its MVC decoding engine has the same bug than CoreAVCDecoder (and also to test if a badly authored Ice Age 3 3DBD could be the culprit). Since DVDFab can correctly decode the MVC stream, I can only deduce that there is a bug in CoreAVCDecoder. However, there must be something strange or new in the Ice Age 3 BD, as all encodings of other 3DBDs I did were successful.

[r0lZ]

Oh, well, the decoding of Ice Age 3 with the Intel SDK is even worse:



However, I did another quick test with the NetBlender demo 3DBD (that has no problem with the CoreAVCDecoder), and it worked perfectly. So, that confirms that there is something strange (or a new variant of the MVC encoding?) in Ice Age 3, but apparently, the SDK decoder is as good as the CoreAVCDecoder for the vast majority of the 3DBDs currently available.

If you are interested in trying it, here is what you should do.
I had to download and install the RawSource Avisynth plugin.

1. Combine the AVC and MVC streams to H264 with MVCCombine. (Do not mux to M2TS.)
2. Convert the Combined.h264 file to 2 YUV files with the sample_decode_x64.exe from the SDK. (You need much disc space!):

sample_decode_x64.exe mvc -i Combined.h264 -o TEST

3. Use this script (or similar) to encode in SBS:

LoadPlugin("rawsource.dll")
left =RawSource("TEST_0.yuv", 1920, 1080, "I420")
right=RawSource("TEST_1.yuv", 1920, 1080, "I420")
StackHorizontal(left, right)
BilinearResize(1920, 1080)

No problem at all with the NetBlender demo ISO, but as you can see above, that doesn't work well with the Ice Age 3 BD. (In addition, Ice Age 3 has the left and right view inverted, but that's not a problem.)

Anyway, that confirms that the Intel SDK could be a good solution for us. With some coding, it should be possible to convert it to an avisynth plugin that will work with most BDs, without the need to produce large intermediate files, and (probably) without seek problems. It should even be possible to encode directly from the BD without having to write any file on disc, with the help of pipes.
We can also hope that the "Ice Age 3 bug" will be fixed by Intel. If it's not the case, we have the source code, and we can try to find and fix the bug. (Unfortunately, I can't help on that point. I'm not a C++ coder, and I know nothing in decoder programming.)

I wonder what makes Ice Age 3 so difficult to decode. Someone has an idea?

[jj666]

Originally Posted by r0lZ

I wonder what makes Ice Age 3 so difficult to decode. Someone has an idea?

I have the Ice Age 3 disk that was bundled with the Panasonic players - this is also not processable in Scenarist (I forget which error but a missing flag in the H264 stream or similar). Is this also the version you were trying r0lz? Or a store bought Ice Age 3, which I assume would be a different encode.

Cheers,

-jj-

[Thalyn]

Slavanap: Given your newest build a bit of a run, but I'll be damned if I can get the auto-detection to work properly. I'm not sure whether it's just not working right with branched movies but I've had all sorts of results from around 100 frames less than there should be to almost 200 more. Not to mention it takes a lot longer to suss it out than DirectShowSource (not the MVC variant) does - many minutes against a second at most, even with only a single source file. This has subsequently limited my testing also of branched sources.

On the plus side, it's working fabulously with single-file sources whether they're CombineMVC output or raw SSIF files; provided I copy them off the BluRay first. I tried going straight from the disc once... let's just say that I could have copied it to the harddrive and run the transcode three times over with a couple hours to spare in the time it would have taken to go direct from the disc just once.


I do like the sounds of that Intel decoder, though - and the encoder as well, if I'm honest. It would be nice not having to jump through hoops and having an MVC encoder would also mean being able to store my 3D movies in more backwards-compatible format. Frame Sequential might get the same or better compression but if you play it back through something that's not FS-aware than you're not going to have a good time, where MVC is transparent if it's unrecognised.

Definitely some promise there.

A thought on Ice Age, though: Monsters, Inc gave me some weird errors if I worked from the SSIF files (sync issues, missing frames, etc). However, if I worked from the M2TS files, even though they're theoretically the same stuff, it was fine. I had to use MakeMKV to find out what files made up the MVC encode since I don't know how to read playlists yet. Perhaps something you could look into for Ice Age 3?

[jdobbs]

Originally Posted by jdobbs

@slavanap

I'm running the newer version on a job with a huge number of segments right now... it's working great, with an impressive speed improvement over the last release. Sweet. I'll let you know how it works out.

I've now done four titles with this version, including one that froze on the previous release. All have completed successfully and I've seen no issues. I haven't tried the automatic framecounts at all, since I get that information easily from the MPLS. In fact, this release has worked so well that I'm considering using it (replacing DirectshowMVCSource) in the next BD Rebuilder release. I need to do more testing, though.

Thanks.