DGMPGDecNV 1.0.2

[Guest]

Originally Posted by Turtleggjp

I'm having trouble with the hardware bob deinterlacer.

Try both field orders with the use_top_field set as true and false. If it won't work either way then upload a failing source stream.

[hajj_3]

Originally Posted by neuron2

I have finally started it. I have DGMPGIndexNV working with video and AC3 and MPA audio. I plan to get DGDecodeNV working next and then add the other audio types. What audio types do you need most?

I don't plan to support MPEG1 video or MPEG1 program streams. Is that a problem?

Thats great i just cap ws.pdtv and 1080i mpeg2 so .ac3 and .mpa/.mp2 support is all i need and pretty much all anyone needs really, not many countries use mpeg2 with aac.

have u done any speed tests comparing it with the normal dgmpgdec? 1080i takes a while to make a .d2v file, SD is fairly quick, atleast for short tv shows.

No-one uses mpeg1 anymore, no point adding that.

[Turtleggjp]

Uploading a sample now. It is a .ts file trimmed with HDTV2MPEG 1.11.89.

In the indexer, you can see the repeated frames, but no combed frames (even with Full HD display enabled). If I index the file, and serve it with AVISynth, I can see repeated frames and the following combed frames. If I set deinterlace=1 with use_top_field=true, it looks smooth (no repeated/dropped frames). With use_top_field=false, I can see duplicated frames, followed by a noticeable jump caused by a missing frame.

Here is the sample:
http://www.mediafire.com/?sharekey=f...4e75f6e8ebb871

[halsboss]

I have fast sports field action HD sources with a lot of blocking (the channels mangle these, unfortunately, also upscaled SD due to inter-channel agreements). The non-NV version seems to provide deblocking natively and easily (can I make it stronger ?)

#ipp=true interlaced post-processing 
#cpu=4    DEBLOCK_Y_H, DEBLOCK_Y_V, DEBLOCK_C_H, DEBLOCK_C_V 
#cpu=6    DEBLOCK_Y_H, DEBLOCK_Y_V, DEBLOCK_C_H, DEBLOCK_C_V, DERING_Y, DERING_C 
MPEG2Source("G:\HDTV\q2.d2v",info=0,ipp=true,cpu=6)

Does/will the NV version have the same facility ?

[Guest]

Originally Posted by halsboss

Does/will the NV version have the same facility ?

No, because the GPU delivers the decoded frame and my code has no access to the quants. You can use a deblocker in your script, such as BlindPP(), Deblock(), and others.

[Guest]

1. When demuxing with the CLI, audio filenames no longer erroneously include the string ".dgm" in the filename.

2. Program streams are now supported. Currently, only MPA and AC3 audio is supported.

3. The machine ID field in the About box is now copyable.

4. Fixed a bug in PAT/PMT parsing.

5. Added a warning popup for cases of blank audio at the start of the stream that can cause incorrect reported audio delay. User is recommended to skip GOPs at the beginning of the project.

http://neuron2.net/dgmpgdecnv/dgmpgdecnv.html

Next I want to add pulldown handling and address the Purevideo deinterlacing issues that have been raised.

[hajj_3]

ooooo

The license file would work with DGAVCDecNV too right?

How is the purevideo de-interlacing doing, does it do a good visual job, is it better than leakkerneldeint or field de-interlace (no blend)? Is it faster than those?

If i wanted to use this with gordianknot what would i need to put in the .avs to be able to load the .dgm file and would there be any way for me to check the cropping in gordian knot or would i have to use a different app to find out the cropping then manually enter the cropping needed into a .avs file?

keep up the great work neuron!

[shon3i]

Something wrong when i select whole DVD or multiple vobs. Decoding stoped after first vob.

[Guest]

Read the release notes.

[halsboss]

Originally Posted by neuron2

2. Program streams are now supported. Currently, only MPA and AC3 audio is supported.
etc

Nice. Thank you very much Neuron2.

[halsboss]

A hopeful query: if the GPU is doing the work of decoding and returning frames, and can deinterlace if you ask it to, are the cards/cuda smart enough to re-size (interlaced/deinterlaced) frames for you too to a specified size ? I guess I'm asking if that's possible, then would the NV based decoders may be able to resize something like HD1080i->SD576i. That'd be really really handy, if it could I ask about interlaced in particular since a lot of my footage is fast sports field action where the interlaced fields have quite large differences.

[squid_80]

Actually they can, the API allows the program to set the desired size of the decoded frames.

[Guest]

Originally Posted by squid_80

Actually they can, the API allows the program to set the desired size of the decoded frames.

Can you be more specific and/or provide sample code, please?

[squid_80]

From the cudaVideoDecode SDK sample app:

VideoDecoder::VideoDecoder(const CUVIDEOFORMAT & rVideoFormat)
{
    // Validate video format. Currently only a subset is 
    // supported via the cuvid API.
    cudaVideoCodec eCodec = rVideoFormat.codec;
    assert(cudaVideoCodec_MPEG1 == eCodec || cudaVideoCodec_MPEG2 == eCodec || cudaVideoCodec_H264 == eCodec);
    assert(cudaVideoChromaFormat_420 == rVideoFormat.chroma_format);
    // Fill the decoder-create-info struct from the given video-format struct.
    memset(&oVideoDecodeCreateInfo_, 0, sizeof(CUVIDDECODECREATEINFO));
    // Create video decoder
    oVideoDecodeCreateInfo_.CodecType           = rVideoFormat.codec;
    oVideoDecodeCreateInfo_.ulWidth             = rVideoFormat.coded_width;
    oVideoDecodeCreateInfo_.ulHeight            = rVideoFormat.coded_height;
    oVideoDecodeCreateInfo_.ulNumDecodeSurfaces = FrameQueue::cnMaximumSize;
    // Limit decode memory to 24MB (16M pixels at 4:2:0 = 24M bytes)
    while (oVideoDecodeCreateInfo_.ulNumDecodeSurfaces * rVideoFormat.coded_width * rVideoFormat.coded_height > 16*1024*1024)
    {
        oVideoDecodeCreateInfo_.ulNumDecodeSurfaces--;
    }
    oVideoDecodeCreateInfo_.ChromaFormat        = rVideoFormat.chroma_format;
    oVideoDecodeCreateInfo_.OutputFormat        = cudaVideoSurfaceFormat_NV12;
    oVideoDecodeCreateInfo_.DeinterlaceMode     = (rVideoFormat.progressive_sequence) ? cudaVideoDeinterlaceMode_Weave : cudaVideoDeinterlaceMode_Adaptive;
    // No scaling
    oVideoDecodeCreateInfo_.ulTargetWidth       = oVideoDecodeCreateInfo_.ulWidth; // Set this to the desired output width 
    oVideoDecodeCreateInfo_.ulTargetHeight      = oVideoDecodeCreateInfo_.ulHeight; // Set this to the desired output height 
    oVideoDecodeCreateInfo_.ulNumOutputSurfaces = 2;  // We won't simultaneously map more than 2 surfaces
    
    // create the decoder
    CUresult oResult = cuvidCreateDecoder(&oDecoder_, &oVideoDecodeCreateInfo_);
    assert(CUDA_SUCCESS == oResult);
}

I haven't played with different values myself so YMMV.

[halsboss]

this does interlaced resizing, right ? sorry to harp on it, but that's important for some people (well, me) with 1080i captured sources with lots of movement and panning/zooming, and only a SD (PAL 576) DVD player to watch it with)

[Guest]

Hold your horses there, bro'. We don't even know if it resizes at all yet.

[halsboss]

No worries I hope, but do not expect at all ! Imagine it, a GPU based sourcer/resizer. (I know, there's probably heaps of pragmatic issues which would rule it out). I wish upon a star that Nvidia had deblocking built-in as well, but that's for another decade or century All the best.

[Turtleggjp]

Originally Posted by neuron2

Next I want to add pulldown handling

Sweet. I'm unable to use your NV version for working with my TV shows right now, as most of them are using pulldown.

Originally Posted by neuron2

and address the Purevideo deinterlacing issues that have been raised.

Cool, so I can assume that you were able to repeat my problem on your end?

[Guest]

Yes, and Nvidia is looking at it. It appears to affect MPEG2 only.

[laserfan]

Hey I notice you've slipped-in a newer DGDecodeNV.dll to this release package! I'm gonna assume I should use this latest version for my VC1 and AVC testing now, unless you tell me differently...

EDIT: Oops--now I looked at the Properties of the old vs. new and they appear identical anyway! Never mind then!