Authors:
(C)2004 Alexander G. Balakhnin aka Fizick,
(C)2005 Turyst04, [email protected] - synchronous modes and audio info using
VersionVersion 0.96
Download: http://avisynth.org.ru
Category: Telecine
Requirements: Avisynth 2.6
GetDups - is a very special plugin for selecting unique duplicate frames from video clip. License GPL v2.
It was developed for processing of special clips, produced by video capturing of old 8 mm (Super 8) films projected with projector without (removed) rotary shutter.
Such source clip contains groups of original frame duplicates (very similar), separated by transitional smeared (not-similar) frames. The plugin analyses amd selects one good original frame per such group and skips all others to restore original film frames (sort of telecine). The plugin support two capturing modes: nonsynchronous at low projection velocity (below 8 fps) and synchronous with stabilized projection velocity 16,67 fps (20 fps for NTSC world). It also has a special audio mode for using clicks of claw intermittent for frame pause detection.
GetDups( clip, int "threshold", int "fade", bool "chroma", bool "tff", int "mode", int "show", int "log", string "logfname", int "audio", int "audio_thr", int "audio_shift", bool "debug", int threshold2, int "diff", clip "mc")
first parameter - input clip
threshold - threshold of difference between similar neighbor frames (real percent, from 0.0 to 100.0, default= 7.0)
fade - maximum of luma correction (default = 10)
chroma - use chroma information too (default = false ), now disabled
tff - fields order - top fields first (=true) or bottom (=false) (default = as Avisynth assumes)
mode - processing mode (-1, 0, 1, 2 or 4) depending on input capturing mode:
0 - nonsynchronous (old mode, only one up to the version 0.5 inclusive) - first similar (by threshold) duplicate is selected after different (by threshold2) transient frame ;
1 - synchronous mode (16,67 fps), get on output 2 neighbor fields (from 3 input);
2 - synchronous mode (16,67 fps), get on output 1 field only (from 3 input);
4 - nonsynchronous mode (like mode=0) with temporal smoothing of difference values;
-1 - nonsynchronous mode (like mode=0) with last duplicate selected ;
-2 - nonsynchronous mode (like mode=0) with best (most similar to previous) duplicate selected ;
(default = 0).
show - mode of processing and info show
(0 - processing, 1 - processing and info, 2 - test and info, default = 0)
log - log mode:
0 - without Log (only one up to the version 0.7 inclusive);
5 - save source frame numbers to log-file;
1 to 4 - save statistic information about frame difference depending on output format or loudness params (for audio=1 in 3 columns: min, max and power) for each frame:
1 -Log-file contains difference between all neighboring fields - Top with Bottom, then Bottom with Top of next frame, 1 column,
2 - Log-file contains difference between all fields - Top with Top, Bottom with Bottom, of neighboring frames, 2 columns - 1-st for Top, 2-nd for Bottom (if mode=4 Log-file contains temporal smoothing difference values),
3 - Log-file contains raw values (without smoothing) of difference between all fields - Top with Top, Bottom with Bottom, of neighboring frames, 2 columns - 1-st for Top, 2-nd for Bottom,
4 - Log-file contains both raw and smoothed values of difference between all fields - Top with Top, Bottom with Bottom, of neighboring frames, 4 columns - 1-st for Top, 2-nd for Bottom (raw), then again Top and Bottom (smoothed),
default = 0.
logfname - save Log in specified file name and path (default = "c:\\gd_stat.log")
audio - audio info using mode:
0 - not used audio info (only one up to the version 0.8 inclusive),
1 - use audio info (detect "good frame" without video info analyze),
2 - use audio info (like audio=1, but force protecting from misoperation in case of intermittent mechanism 2-splashed sound)
(default = 0).
audio_thr - threshold of audio absolute value: from 0 to 32768 (default=5000); best value mast lay between lowest pick and highest bottom values in all of clip audio (use show mode for analyze).
audio_shift - shift from default detected "good frame" - positive or negative value (reasonable values for projection velocity 4-8 fps is from -2 to 2), if default value audio_shift=0 make bad results, try tune this param.
debug - output info for DebugView utility (default=false)
threshold2 - second (high) threshold of difference between non-similar (transional) frames, suggested value is about threshold*5, default=threshold (for compatibility with old versions with single threshold).
diff - kind of inter-frames difference.
0 - max SAD of 8x8 block (default), 1 - mean SAD of frame (less sensible to hair movement).
MC - motion compensated clip for previous (compensated) frame comparing with current source frame. Default is source clip.
1. Works only YUY2, YV12 and any other planar (since v0.9.4) color format.
2. Developed for interlaced video cameras but works well for progresive too.
3. Tested with Avisynth 2.6.0.
4. No arbitrary access to frames. Plugin correctly works only with strictly sequential processing from first frame to end of clip.
5. Uses method and code of pixels block comparing from Dup plugin (by Donald Graft/Klaus Post).
6. Since version 0.2, plugin uses not inter-frame differences, but inter-fields (for correspondent fields of neighbors frames). It outputs a frame that have a second field chronologically between these duplicate fields. Correct fields order definition (tff, bff) is very important (for interlaced video).
7. Since version 0.6, synchronous mode is added (by Turyst04) for work with input clip, captured by PAL camcorder with stabilized projected velocity 16,67 fps, and suitable mode parameter. Input clip must be processed as separated by fields. This mode allows to increase the capturing velocity (and to minimize the size of input clip file) without any loss of output quality. You can also try use modern 50 fps progressive cameras (not tested).
8. Since version 0.9, audio info using for alternative finding method mode is added (by Turyst04, partially used code from AudGraph plugin by Richard Ling) for work with input clip, captured with clear pawl intermittent mechanism (greifer) sound; suitable params added: audio, audio_thr, audio_shift.
1) for nonsynchronous mode:
Avisource("input.avi")
loadplugin("GetDups.dll")
AssumeTFF()
fc=Framecount()
GetDups(mode=-2,threshold=2.0)
AssumeFps(16)
Trim(0,fc/3)
2) for synchronous mode:
Avisource("input.avi")
loadplugin("GetDups.dll")
fc=Framecount()
AssumeTFF()
SeparateFields()
GetDups(mode=1)
Weave(AssumeFieldBased)
AssumeFps(16)
Trim(0,fc*2/3)
3) for audio info using mode:
Avisource("input.avi")
loadplugin("GetDups.dll")
AssumeTFF()
GetDups(audio=1)
AssumeFps(16)
#Trim(0,999)
Plugin is for special method for 8 mm films capturing by camcorder (not for simple direct shooting from screen, not for per-frame one-shot capturing).
1) modify projector apparatus by removing of rotary shutter to prevent brightness variations;
2) set low projection frequency about 4-6 fps (nonsynchronous mode) or stabilizing velocity 16,67 fps (synchronous mode);
3) project to small screen, shoot and capture video clip to file;
4) review clip by plugin GetDups in test mode (show=2, may be with Log), check duplicates existence and detection.
If results are bad, then analyze Lof-file, tune threshold, check tff, try mode=4, try use audio=1 or re-capture film with lower fps .
5) make normal processing, write result to output file. Set output clip fps to 16 or 18 by AssumeFps.
6) use only selected unique frames at beginning of clip as final video, remove rest frames by trim.
The some (currently small, please post more) discussion is at forum
http://forum.doom9.org/showthread.php?s=&threadid=73577
Great discussion about capturing of 8 mm (Super 8) films and GetDups plugin at Russian forum: http://forum.ixbt.com/0029/016088-8.html#234
How to make stabilizing velocity 16,67 fps projector and other aspects of synchronous mode - look this supplement article (by Turyst04, in Russian only).
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
Documentation is distributed under CreativeCommons BY-SA 3.0 license.
Please consider to make some donation to be registered user.
Version 0.1, November 2, 2004 - first public. Thanks to Eugen65 for plugin idea and discussions.
Version 0.2, November 3, 2004 - Fields analysis method, mean luma correction, Show parameter now integer, ignoring of drops in captured frames.
Version 0.3, November 7, 2004 - Different frames now defined by one filed (was by both field), final frames now is copy of last duplicate.
Version 0.4, November 8, 2004 - Duplicate is now selected by most similar field if both difference are below threshold.
Version 0.5, November 8, 2004 - Compared block size decreased to 8 instead of 32.
Version 0.6, March 13, 2005 (Turyst04), version for debug only, isn`t published - added synchronous mode and suitable mode parameter for work with input clip, capturing with stabilizing velocity 16,67 fps.
Version 0.7, March 23, 2005 (Turyst04) -changed output fields finding method in mode=1 and mode=2:
threshold is not used (in contrast to mode=0), simply looking for local min and max fields differential values (fields with minimal differences is used foroutput).
Version 0.8, April 10, 2005. Changed by Turyst04: added output Log (for external analysis) and mode=4 for temporal smoothing method (smooth difference values only, not picture) combined with old method (with threshold, like in mode=0).
Changed by Fizick: corrected mode=0 algorithm: now plugin does NOT detect the frame as a (false) dup, if the small difference is found for chronological first field only, and the same field difference was small at previous step too.
Version 0.5.2, April 6, 2005 - was not public, released July 6, 2005 mostly for historical purposes :-). This special version is based on v.0.5 and contains my (Fizick) contribition to v.0.8 (must produce the same output as v.0.8 in mode=0) - Corrected mode=0 algorithm: now plugin does NOT detect the frame as a (false) dup, if the small difference is found for chronological first field only, and the same field difference was small at previous step too.
Version 0.9, June 21, 2005 (Turyst04) - added audio info using for alternative finding method audio=1:
- added params audio, audio_thr, audio_shift.
Version 0.91, September 20, 2005 (Turyst04) - fixed some bugs, added method audio=2
Version 0.92, March 13, 2014 (Fizick)
- fixed v.9 implicit TFF bug;
- added log=5 mode as requested by Dan
Version 0.93, October 25, 2015 (Fizick)
- added mode=-1 with lst duplicate
Version 0.94, March 3, 2016 (Fizick)
- added threshold2, diff and MC parameters. Aviynth 2.6.0 with all planar support.
Version 0.95, May 3, 2016 (Fizick)
- changed default threshold2 value from threshold*5 to threshold for compatibility with older versions up to 0.93,
, - fixed shown frame numbers in show=1 mode,
- alowed single different frame in mode=-1 mode (like v0.93).
Version 0.96, May 8, 2016 (Fizick)
- added mode=-2 with best (most similar) duplicate.