Re: [Mjpeg-users] Re: Encoding large DV to MPEG-2

[Steven M. S. <sm...@2B...>]

On Fri, 9 Jan 2004, George Kola wrote:

>         That was a problem. I found that smil-utils-0.1.3 was released. I 
> tried that and had problem with reading after  62% of the DV file. I picked 

	Interesting.   What was the nature of the problem?  Did smil2yuv
	coredump or declare the file to be somehow broken?

> the version from CVS and it worked like a charm. Thanks for the suggestion.
	
	Ah, super!  It has been a while since 0.1.3 was released and there
	have been numerous changes/fixes since then (as well as several new
	utilities and filters).

>        I was using libdv-0.99. I now picked up the latest from CVS.

	Great.   The libdv project does not release often so I have used the
	CVS version for a long time (same thing for Kino).

> > checking libavcodec version (build >4644)... yes
>            
>          Originally, it was not finding it. Now I pointed it correctly 
> and I get the line as above and it uses libavcodec.

	Wonderful.   

> I am able to generate VCD successfully. But I am having problem with mpeg-2. 
> I am using 
> 
> smil2yuv -i 2 ../test.avi 2>smil.err | yuvdenoise -S 0 -l1 2>yuv.err | 
> y4mscaler -O chromass=420_MPEG2  2>y4m.err |mpeg2enc -f 8 -M 2 -E -10 -2 1 -q 
> 6 -K kvcd -o test.m2v >mpeg2.out 2>mpeg2.err

	Good choice of parameters.   If the filesize is not too large you 
	can use the TMPGEnc tables ("-K tmpgenc") for slightly higher 
	quality.

> (test.avi is a 13GB ODML DV file)
> 
> I find that, the final mpeg2 generated contains only 16.39 minutes of video. 

	I'm curious - how was the playing time measured - with 'mplayer'?

> I generated the wav using smil2yuv -a test.wav and the audio is fine -- full 

	it is also possible to do both the video and audio at the same time -
	just add the "-a test.wav" to the command:

	   smil2yuv -a test.wav -i 2 ...

> 1 hour. Similarly the MPEG-1 VCD is full 1 hour. Only the mpeg2 video is not 
> fine. I am at a loss. If I dump the output of smil2yuv to disk, I get a 52 GB 
> file (DV file is 13 GB), so that seems about right. None of the tools report 
> any error. 

	The .y4m file was created with this I presume:

smil2yuv -i 2 ../test.avi |yuvdenoise -S 0 -l1 |y4mscaler -O chromass=420_MPEG2
 > test.y4m

> [fireplace:mpeg2]$tail -3 y4m.err
>    INFO: [y4mscaler] Frame number 107913
>    INFO: [y4mscaler] Frame number 107914
>    INFO: [y4mscaler] End of stream at frame 107914.
>           
> [fireplace:mpeg2]$tail -7 mpeg2.err
>    INFO: [mpeg2enc] Frame end 107910 B quant=6.00 total act=2769634.95788
>    INFO: [mpeg2enc] Frame end 107911 I quant=6.00 total act=2769810.90260
>    INFO: [mpeg2enc] GOP start (0 frames)
>    INFO: [mpeg2enc] Frame end 107912 B quant=6.00 total act=2769827.68565
>    INFO: [mpeg2enc] Frame end 107913 B quant=6.00 total act=2769844.56189
>    INFO: [mpeg2enc] Guesstimated final muxed size = 803803550

>            I think mpeg2enc may be the problem. I do not know if I need 
> to pass any other option. The MPEG-2 is for archival purpose that 

	But mpeg2enc has logged that it encoded all 107913 frames.  What does
	"ls -l test.m2v" say?   mpeg2enc's estimate is 803803550 (which
	includes a 224k audio stream unless -B has been used).

	No other options are required that I can think of.

	What happens if you do something like this with the 52GB file:

mpeg2enc -f 8 -M 2 -E -10 -2 1 -q 6 -K kvcd -o testing.m2v < BIGFILE.y4m

	Is the resulting file still only ~16 minutes?

>       I have tried just passing output of smil2yuv to mpeg2enc and got the 
> file with the same size.

	And that size is? ;)

	Another thing to try is:

	   smil2yuv test.avi |yuvplay -c

	and see if the entire video portion of the movie is being output by
	smil2yuv.   Without the -c that would take ~1hr but with the -c
	it will play as fast as your system can decode/display the data.

>        For the successful VCD, I used 
> 
> smil2yuv test.avi | yuvscaler -O VCD | mpeg2enc -f 1  -o test.m1v

	Another method would be "y4mscaler -S option=sinc:4 -O present=VCD", the
	sinc scaling kernel can give better looking output.

	Out of curiosity what system are you doing this on?

> p/s: Thanks Steven, for helping me out.

	Welcome - hope the remaining issues can be resolved, at least some
	progress has been made!

	Cheers,
	Steven Schultz