summaryrefslogtreecommitdiffstats
path: root/debian/transcode/transcode-1.1.7/docs/README.filter.modfps
blob: 40a6c75e54115e206e57fd82423a250bb545eb57 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324

I. Introduction

This file contains some example commands, and scripts which I've
found useful when working with modfps.  Additionally there's some
general documentation on how to use it.

modfps is a filter written for transcode which allows one to alter
the frame rate by either dropping or cloning frames.  While one might
initially think that there's a better more intelligent way to alter
the frame rate, most things I've toyed with which seriously compromise
the original frames look poor.

2. Options

first of all, I expect that you've run "tcmodinfo -i modfps" ... that's
a good way to at least find out about the options.  I'll try giving
some more explanation for each option.

infps, infrc are depreciated, and I can't immediately think why they
might be used.

verbose: for each frame, some information about the time the frames
	sit in the stream, and whether a frame is skipped/cloned will
	be printed.  This is info is probably not useful except for
	debugging.

mode :	There are two modes of operation, buffered (1) and fast (0).
	Buffered mode is the default and recommended mode of operation.
	While fast mode will get you the correct frame rate, depending
	on the video stream you're working with it may look very poor

	For example, if you have a 29.97fps stream recorded from your
	tv card, I've frequently found that it's really 23.976fps or
	25.0fps, and there are just duplicate frames.  If you use fast
	mode to convert the steam, your odds are one in 5 if you're
	going from 29.97 to 23.976 of randomly hitting the pattern
	such that the skipped frames are the duplicate frames.  If you
	miss, you'll skip a frame with action, which gives you a big
	jump, and soon after that they'll be the duplicate frame.  This
	gives one a bad headache if you try to watch to much of it on
	a monitor (I'm not sure if it might be acceptable output to
	a tv.

	Unless you really need the speed, I'd suggest sticking with
	mode 1.

buffer: The number of frames to buffer.  The default is 5, which is
	perfect when going from 29.97fps to 23.976.  As a note, while
	it might initially seem like a good idea to run with 100 buffers
	if you've got the RAM, note that modfps does not consider local
	effects when deciding which frame to skip/clone.  Thus, if one
	were going from 29.97fps to 15fps and there were 60 frames
	buffered, it could decide that all of the frames that it wants to
	skip are the last ones, which would mean that AV could be off
	by 1.5 seconds.

	A good formulae to decide on the number of buffers to use is:
	buffers = absolute value of ( input_fps/(input_fps-output_fps) )

subsample: When calculating the difference of frames, how many pixels do
	we look at?  If we looked at all of them, buffered mode would
	run slower, while not behaving too differently.  The current
	default is to run with subsample=32 , or to say in English, to
	compare every 32nd pixel.  on a 640x480 frame this gives 9600
	points of comparison. ... on the 640x240 frames I record from
	tv, it's 4800 points of comparison, and I've yet to see modfps
	choose a bad frame to skip, but perhaps your source is different
	enough to warrant setting this lower.

clonetype: Ah, the fun stuff.  When cloning a frame, modfps will let
	one do more than a simple copy of the data.  Suppose that the
	first cloned frame is frame 5.  Frame 5 and 6 of the import
	steam will correspond to frame 5 and 7 of the export steam.
	Frame 6 of the export steam will varry based on clonetype.

	0: no special behavior, export frame 6 will be the same as
	export frame 5.

	1: export frame 6 will be composed by alternating the rows
	from export frame 5 and 7, with the top row being from frame
	5 .  If the display device is interlaced, like a TV, this
	might look best (it's similar to how a 23.976 fps dvd is
	displayed at 29.97 fps; some of the frames have alternating
	rows from the frame previous and after).

	2: export frame 6 will be composed by alternating the rows
	from export frame 5 and 7, with the top row being from frame
	7 .  This is similar to value 1, except the order of the
	frames is switched.  I'm not sure if this would be useful, but
	it was trivial to add.

	3: every pixel of frame 6 will be an arithmetic average of
	the corresponding pixel from frame 5 and 7.

	4: First we examine where the cloned frame sits in the export
	stream, and compare it to where it came from in the import
	steam.  some of the time, the following frame is closer to
	"correct" based on timestamps.  If so, frame 6 is a copy of
	frame 7.

	in other cases, the frame might be temporally weighted closer
	to frame 5, or between 7.  In such cases, we assign a weight
	to frame 5 and 7 such that the sum of the weights are 1.
	Then each pixel of frame 6 will be the sum of the weight of
	the frame times the pixel of the frame.  I.E. if the temporal
	weight is closest to frame 5, frame 6 will look mostly like
	frame 5, but will have some info from frame 7 in it.

	I think for non-interlaced displays this or value 5 will give
	the best results.

	5: for this, I considered how a TV looks, with the bright
	pixels "sticking around" better than the less bright areas.
	This is currently only implemented for YUV (and it will
	probably stay that way).

	An approximate ordering of the speed of the different clone
	types is: 0, 1, 2, 3, 4, 5 from fastest to slowest.   In my
	opinion, modes 0-4 are fast enough to not worry about speed,
	while mode 5 is slow enough that you should test and make
	sure you like the results before doing a lot with it.


3. Misc.

Because transcode by default will buffer some (5) frames, whenever
one isn't encoding an entire file (I.E. using "-c" "-J cut" or
"-J skip").

As of transcode 0.6.9 filters do not see frames which are not in
the range of "-c" ... thus if one is cutting frames 1000-1500 the
first frame modfps sees is 1000 .  As modfps buffers the frame it
sets the TC_FRAME_IS_SKIPPED flag, so it won't be encoded.  After
it's read in the buffers it wants, it will start outputting the
frames it's buffered.  I.E. the first frame modfps doesn't set as
skipped will be 1005 from the encoder's point of view while
corresponding to frame 1000 from the import stream.

When using -c ,  one should add the number of buffers to the last
segment, because all of the frames are offset by a bit.

I.E. if: "-c 3916-13090,17493-24185,28304-42212,46777-54710" gives
you the segments you want without modifying the frame rates, when
you use modfps=buffers=5 (the default) you should modify the option
to "-c 3916-13090,17493-24185,28304-42212,46777-54715"

Also of interest is that ivtc works as well with modfps as it
does with decimate.  While there's no real benefit to use it when
transcoding from 29.97 to 23.976 fps I've seen 29.97 video which has
telecine effects, except it's at 25.0 fps ... "-J ivtc,modfps=buffers=7"
works well in this case.

When using -J cut, cut will alter the number of audio frames skipped
so that "-J cut=1000-1500,modfps -f 29.97 --export_fps 23,1" will keep
the AV in sync while -c wouldn't.  However, again all the frames in
the export are off by the number of buffers.  But the effects are
different from -c , so if you wanted to use
"-J cut="3916-13090,17493-24185,28304-42212" you would then use
"-J cut="3921-13095,17498-24190,28309-42217"

4. Example commands

the source video of the example was recorded from my tv card, with
half the height of the source (to avoid interlacing effects, and
because I can't encode 640x480 realtime) .  Because of this, the
video I get is usually either 29.97fps (10%) 23.976 (65%) 25.0 (25%).
The percentages correspond to roughly how often I come upon this
source.  As you can see, I rarely have something that will look
good without modification, and even then, decimate wouldn't always
save me.

First, I do a first pass of xvid, video only.

/usr/bin/nice -n +19 /usr/local/test/bin/transcode --print_status \
5 --export_fps 25.0,3 -x mplayer -J modfps=buffer=7,hqdn3d -V -i tv.avi \
-b 64 -c 3916-13090,17493-24185,28304-42212,46777-54715 -y xvid,null \
-j 0,8 -Y 0,-4 -B 0,39,08 -w 600 -R 1

Next, I do the 2nd pass, again, video only.

/usr/bin/nice -n +19 /usr/local/test/bin/transcode --print_status 5 \
--export_fps 25.0,3 -x mplayer -J modfps=buffer=7,hqdn3d -V -i tv.avi \
-b 64 -c 3916-13090,17493-24185,28304-42212,46777-54715 -y xvid,null \
-j 0,8 -Y 0,-4 -B 0,39,08 -w 600 -R 2 -o tmp.tv.avi

Now that I have the video, I examine audio.  I've found that since
commercials are much louder than TV, before I give anything to filter
astat, I'll rip the raw audio.  Note, that the values of -c are
different for audio than they were for video.

I wish to stress: Currently, one cannot use "-c" with transcode and
process audio and video at the same time while changing the frame
rate without losing AV sync.

/usr/bin/nice -n +19 /usr/local/test/bin/transcode --print_status \
500 -f 29.970 --export_fps 25.000 -u 35 -x null,auto -y null,wav -m \
sound.tv.avi -c 3267-10919,14593-20175,23611-35212,39020-45637 -i tv.avi

Then we do a quick pass thru with filter astat.

/usr/bin/nice -n +19 /usr/local/test/bin/transcode --print_status 500 \
-f 23.976,1 -p sound.tv.avi -J astat=alog -u 50

Finally, I encode the audio while using passthru mode to merge it with
video.

/usr/bin/nice -n +19 /usr/local/test/bin/transcode --print_status 50 \
-i tmp.tv.avi -o out.tv.avi -p sound.tv.avi -s `cat alog` -P1 -b 64 \
-y raw

Please note in the above example that modfps was only used to handle
changing the rate of the video, and that one must give different
values to -c (as of transcode 0.6.9)

5. Possibly useful scripts

The following scripts are what I use to calculate my cut sections
which are given to transcode.  I use avidemux v2 on input avi's
to cut out commercials from a video stream I had recorded on tv.
I could use avidemux to save the avi, however some of the sections
well be reencodded, and it's a semi-slow process (taking a few
minutes, meaning that it's highly interruptive) so I prefer to just
have transcode take a bit longer to skip over the commercials.
instead I save the edit list, and have my encoding scripts cat
the edit list thru two commands.  The first gives the "-c" values
for video, and the 2nd handles it for audio.

----- BEGIN PERL SCRIPT -----
#!/usr/bin/perl

$numbuffers = 5;
$i = 0;
$argc = @ARGV;

# we take one argument: the number of buffers
if ($argv > 0){
  $numbuffers = $ARGV[0];
}

while(<STDIN>){
  if(/^Start : (\d+)/){
    # even tho initially some frames will be skipped, with newest
    # transcode, we keep the beginning where it should be, but we must
    # add 5 frames to the very last segment.
    $c[$i]{start} = 1 + $1;
    $_ = <STDIN>;
    if(/^Size : (\d+)/){
      $c[$i]{end} = $c[$i]{start}+$1;
      $i++;
    } else {
      print;
      die "ergh, we were supposed to find the end\n";
    }
  }
}

if($i > 0){
  $c[$i-1]{end} += $numbuffers ;
} else {
  print "-c 0";
  exit -1;
}

print "-c ";
for($j=0;$j<$i;$j++){
  print "$c[$j]{start}-$c[$j]{end}";
  if($i-1 != $j){
    print ",";
  }
}
print "\n";
----- END PERL SCRIPT -----

----- BEGIN PERL SCRIPT -----
#!/usr/bin/perl

# read from stdin the workbench

$infps = 5;
$outfps = 4;
$argc=@ARGV;
if ($argc>0){
  $infps=$ARGV[0];
}
if ($argc>1){
  $outfps = $ARGV[1];
}

use POSIX;
$i = 0;
while(<STDIN>){
  if(/^Start : (\d+)/){
    $c[$i]{start} = ceil((1+$1)*($outfps/$infps));
    $_ = <STDIN>;
    if(/^Size : (\d+)/){
      $c[$i]{end} = floor($c[$i]{start}+($1*$outfps/$infps));
      $i++;
    } else {
      print;
      die "ergh, we were supposed to find the end\n";
    }
  }
}
print "-c ";
for($j=0;$j<$i;$j++){
  print "$c[$j]{start}-$c[$j]{end}";
  if($i-1 != $j){
    print ",";
  }
}
print "\n";
----- END PERL SCRIPT -----

The first script takes one argument, the number of buffers
passed to modfps, and will default to 5, like modfps.  The 2nd
script takes two arguments, the import_fps and the export_fps.
Note that the intent of my choice of audio sections was not to
get audio outside of the cutzone (possibly large noise from the
very last bit of a commercial), and thus the video and audio
might end up encoding a different number of frames, and thus
with many cut sections, AV sync might be lost.  However, these
work for me ;)