...es geht wieder los 
neuron2 aka donald coded wiedermal.
...im moment aktuell:
http://neuron2.net/fixd2v/dgmpgdec1013b8.zip
---------------------------------------------------------
DGMPGDecVersion 1.0.13b8 Functional Changes
(versus DGMPGDec 1.0.12)
---------------------------------------------------------
1. A new CLI option for loading input files from a file list contained in a text file
was added as follows:
-Batch-Files -BF ([filelist]) Load batch of files.
The filelist file contains the list of files to load,
one file per line with no extra white space.
2. Previous fixes for stopping crashes when some corrupted files are
loaded broke proper decoding of files with a single I frame (stills), and
files with just a few frames. Those changes were backed out because it
is more important to get normal decoding of correct files working properly.
Nevertheless, I still want to properly address the crashing issue for
corrupt and truncated files, so please notify me if you run into
streams that cause DGDecode to crash at the end of a file. The current
workaround is to remove the last frame entry in the D2V file. My tests
all pass for stills and few-frame files with this change. Please notify me
if you experience any problems in that regard.
3. Some transport files were not being properly decoded due to a missing
check in the transport parser. This is repaired.
4. When quantizer matrix changes occur in a stream, random access would fail
in some cases, producing a corrupted frame. This has been repaired with the caveat
that any file that changes quantizer matrices on the fly must send a sequence
header with every GOP. I have never seen any file that does not do this. Please
notify me if you find one (i.e., one that does not have a sequence header for
every GOP *and* changes quantizer matrices).
5. When processing a transport file with the video PID erroneously set to an
audio stream, DGIndex would declare that the file is not a transport stream
when trying to detect PIDs, and decoding would fail. This has been repaired.
Now a warning box pops up and advises the user to check his PIDs.
6. Fixed a bug in auto-loading from the CLI.
7. Added automatic detection of transport streams. Transport stream files can
now have any file extension.
8. The DC mode decision of postprocess deblocking was broken both horizontally
and vertically. It has been shown to produce objectionable artifacts.
This was repaired by correcting the decision code. (by "tritical")
9. Revised the file open method for the GUI (the CLI remains as before for
backward compatibility). Now, the open file dialog allows multiple selection.
The file names are automatically sorted in an intelligent way. Some further
enhancements may be needed, but I think this is moving in the right direction.
10. Incorporated tritical's changes to DGDecode as follows:
A) added info option to mpeg2source() to display info on frames
- example usage: mpeg2source(info=true)
B) iPP is a bool again. If it is not explicitly set to true or false then it
defaults to auto which switches between field/frame based pp based on the
progressive_frame flag
C) 4:2:2 input support
- can now correctly decode 4:2:2 input sources and correctly output it
to AviSynth as YUY2
*NOTE: mpeg2source() now throws an error on 4:4:4 input, it could handle it
but AviSynth 2.5 does not have internal YUV 4:4:4 support
- 4:2:2 PLANAR post-processing support
D) Upsampling to 4:2:2 from 4:2:0 based on progressive_frame flag. Uses new
upConv parameter of mpeg2source().
- example mpeg2source(upConv=true)
*NOTES: it will only work if input is 4:2:0, if input is anything else
then the upConv parameter is ignored
E) BlindPP now supports YUY2 colorspace
F) other stuff:
- fixed blindPP syntax bug (x and X)
- fixed a small bug with showQ option
- fixed small memory leak with FrameList/GOPList not being free'd
- fixed a bug in the vertical chroma deblocking postprocessing
QP pointer being passed was incorrect for 4:2:0
- faster mmx 4:2:2 to packed YUY2 and YUY2 to planar 4:2:2 conversions
- info output would not work correctly if temporal_reference was not zero
based at the beginning of gops, it does now
[Donald Graft, January 15, 2005]
