1 This document is a tutorial/initiation for writing simple filters in
4 Foreword: just like everything else in FFmpeg, libavfilter is monolithic, which
5 means that it is highly recommended that you submit your filters to the FFmpeg
6 development mailing-list and make sure that they are applied. Otherwise, your filters
7 are likely to have a very short lifetime due to more or less regular internal API
8 changes, and a limited distribution, review, and testing.
13 Let's say you want to write a new simple video filter called "foobar" which
14 takes one frame in input, changes the pixels in whatever fashion you fancy, and
15 outputs the modified frame. The most simple way of doing this is to take a
16 similar filter. We'll pick edgedetect, but any other should do. You can look
17 for others using the `./ffmpeg -v 0 -filters|grep ' V->V '` command.
19 - sed 's/edgedetect/foobar/g;s/EdgeDetect/Foobar/g' libavfilter/vf_edgedetect.c > libavfilter/vf_foobar.c
20 - edit libavfilter/Makefile, and add an entry for "foobar" following the
21 pattern of the other filters.
22 - edit libavfilter/allfilters.c, and add an entry for "foobar" following the
23 pattern of the other filters.
25 - make -j<whatever> ffmpeg
26 - ./ffmpeg -i http://samples.ffmpeg.org/image-samples/lena.pnm -vf foobar foobar.png
27 Note here: you can obviously use a random local image instead of a remote URL.
29 If everything went right, you should get a foobar.png with Lena edge-detected.
31 That's it, your new playground is ready.
33 Some little details about what's going on:
34 libavfilter/allfilters.c:this file is parsed by the configure script, which in turn
35 will define variables for the build system and the C:
37 --- after running configure ---
39 $ grep FOOBAR ffbuild/config.mak
40 CONFIG_FOOBAR_FILTER=yes
41 $ grep FOOBAR config.h
42 #define CONFIG_FOOBAR_FILTER 1
44 CONFIG_FOOBAR_FILTER=yes from the ffbuild/config.mak is later used to enable
45 the filter in libavfilter/Makefile and CONFIG_FOOBAR_FILTER=1 from the config.h
46 will be used for registering the filter in libavfilter/allfilters.c.
51 You now need some theory about the general code layout of a filter. Open your
52 libavfilter/vf_foobar.c. This section will detail the important parts of the
53 code you need to understand before messing with it.
58 First chunk is the copyright. Most filters are LGPL, and we are assuming
59 vf_foobar is as well. We are also assuming vf_foobar is not an edge detector
60 filter, so you can update the boilerplate with your credits.
65 Next chunk is the Doxygen about the file. See https://ffmpeg.org/doxygen/trunk/.
66 Detail here what the filter is, does, and add some references if you feel like
72 Skip the headers and scroll down to the definition of FoobarContext. This is
73 your local state context. It is already filled with 0 when you get it so do not
74 worry about uninitialized reads into this context. This is where you put all
75 "global" information that you need; typically the variables storing the user options.
76 You'll notice the first field "const AVClass *class"; it's the only field you
77 need to keep assuming you have a context. There is some magic you don't need to
78 care about around this field, just let it be (in the first position) for now.
83 Then comes the options array. This is what will define the user accessible
84 options. For example, -vf foobar=mode=colormix:high=0.4:low=0.1. Most options
85 have the following pattern:
86 name, description, offset, type, default value, minimum value, maximum value, flags
88 - name is the option name, keep it simple and lowercase
89 - description are short, in lowercase, without period, and describe what they
90 do, for example "set the foo of the bar"
91 - offset is the offset of the field in your local context, see the OFFSET()
92 macro; the option parser will use that information to fill the fields
93 according to the user input
94 - type is any of AV_OPT_TYPE_* defined in libavutil/opt.h
95 - default value is an union where you pick the appropriate type; "{.dbl=0.3}",
96 "{.i64=0x234}", "{.str=NULL}", ...
97 - min and max values define the range of available values, inclusive
98 - flags are AVOption generic flags. See AV_OPT_FLAG_* definitions
100 When in doubt, just look at the other AVOption definitions all around the codebase,
101 there are tons of examples.
106 AVFILTER_DEFINE_CLASS(foobar) will define a unique foobar_class with some kind
107 of signature referencing the options, etc. which will be referenced in the
108 definition of the AVFilter.
113 At the end of the file, you will find foobar_inputs, foobar_outputs and
114 the AVFilter ff_vf_foobar. Don't forget to update the AVFilter.description with
115 a description of what the filter does, starting with a capitalized letter and
116 ending with a period. You'd better drop the AVFilter.flags entry for now, and
117 re-add them later depending on the capabilities of your filter.
122 Let's now study the common callbacks. Before going into details, note that all
123 these callbacks are explained in details in libavfilter/avfilter.h, so in
124 doubt, refer to the doxy in that file.
129 First one to be called is init(). It's flagged as cold because not called
130 often. Look for "cold" on
131 http://gcc.gnu.org/onlinedocs/gcc/Function-Attributes.html for more
134 As the name suggests, init() is where you eventually initialize and allocate
135 your buffers, pre-compute your data, etc. Note that at this point, your local
136 context already has the user options initialized, but you still haven't any
137 clue about the kind of data input you will get, so this function is often
138 mainly used to sanitize the user options.
140 Some init()s will also define the number of inputs or outputs dynamically
141 according to the user options. A good example of this is the split filter, but
142 we won't cover this here since vf_foobar is just a simple 1:1 filter.
147 Similarly, there is the uninit() callback, doing what the name suggests. Free
148 everything you allocated here.
153 This follows the init() and is used for the format negotiation. Basically
154 you specify here what pixel format(s) (gray, rgb 32, yuv 4:2:0, ...) you accept
155 for your inputs, and what you can output. All pixel formats are defined in
156 libavutil/pixfmt.h. If you don't change the pixel format between the input and
157 the output, you just have to define a pixel formats array and call
158 ff_set_common_formats(). For more complex negotiation, you can refer to other
159 filters such as vf_scale.
164 This callback is not necessary, but you will probably have one or more
165 config_props() anyway. It's not a callback for the filter itself but for its
166 inputs or outputs (they're called "pads" - AVFilterPad - in libavfilter's
169 Inside the input config_props(), you are at a point where you know which pixel
170 format has been picked after query_formats(), and more information such as the
171 video width and height (inlink->{w,h}). So if you need to update your internal
172 context state depending on your input you can do it here. In edgedetect you can
173 see that this callback is used to allocate buffers depending on these
174 information. They will be destroyed in uninit().
176 Inside the output config_props(), you can define what you want to change in the
177 output. Typically, if your filter is going to double the size of the video, you
178 will update outlink->w and outlink->h.
183 This is the callback you are waiting for from the beginning: it is where you
184 process the received frames. Along with the frame, you get the input link from
185 where the frame comes from.
187 static int filter_frame(AVFilterLink *inlink, AVFrame *in) { ... }
189 You can get the filter context through that input link:
191 AVFilterContext *ctx = inlink->dst;
193 Then access your internal state context:
195 FoobarContext *foobar = ctx->priv;
197 And also the output link where you will send your frame when you are done:
199 AVFilterLink *outlink = ctx->outputs[0];
201 Here, we are picking the first output. You can have several, but in our case we
202 only have one since we are in a 1:1 input-output situation.
204 If you want to define a simple pass-through filter, you can just do:
206 return ff_filter_frame(outlink, in);
208 But of course, you probably want to change the data of that frame.
210 This can be done by accessing frame->data[] and frame->linesize[]. Important
211 note here: the width does NOT match the linesize. The linesize is always
212 greater or equal to the width. The padding created should not be changed or
213 even read. Typically, keep in mind that a previous filter in your chain might
214 have altered the frame dimension but not the linesize. Imagine a crop filter
215 that halves the video size: the linesizes won't be changed, just the width.
217 <-------------- linesize ------------------------>
218 +-------------------------------+----------------+ ^
221 | picture | padding | | height
224 +-------------------------------+----------------+ v
225 <----------- width ------------->
227 Before modifying the "in" frame, you have to make sure it is writable, or get a
228 new one. Multiple scenarios are possible here depending on the kind of
229 processing you are doing.
231 Let's say you want to change one pixel depending on multiple pixels (typically
232 the surrounding ones) of the input. In that case, you can't do an in-place
233 processing of the input so you will need to allocate a new frame, with the same
234 properties as the input one, and send that new frame to the next filter:
236 AVFrame *out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
239 return AVERROR(ENOMEM);
241 av_frame_copy_props(out, in);
243 // out->data[...] = foobar(in->data[...])
246 return ff_filter_frame(outlink, out);
251 If you can just alter the input frame, you probably just want to do that
254 av_frame_make_writable(in);
255 // in->data[...] = foobar(in->data[...])
256 return ff_filter_frame(outlink, in);
258 You may wonder why a frame might not be writable. The answer is that for
259 example a previous filter might still own the frame data: imagine a filter
260 prior to yours in the filtergraph that needs to cache the frame. You must not
261 alter that frame, otherwise it will make that previous filter buggy. This is
262 where av_frame_make_writable() helps (it won't have any effect if the frame
263 already is writable).
265 The problem with using av_frame_make_writable() is that in the worst case it
266 will copy the whole input frame before you change it all over again with your
267 filter: if the frame is not writable, av_frame_make_writable() will allocate
268 new buffers, and copy the input frame data. You don't want that, and you can
269 avoid it by just allocating a new buffer if necessary, and process from in to
270 out in your filter, saving the memcpy. Generally, this is done following this
276 if (av_frame_is_writable(in)) {
280 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
283 return AVERROR(ENOMEM);
285 av_frame_copy_props(out, in);
288 // out->data[...] = foobar(in->data[...])
292 return ff_filter_frame(outlink, out);
294 Of course, this will only work if you can do in-place processing. To test if
295 your filter handles well the permissions, you can use the perms filter. For
298 -vf perms=random,foobar
300 Make sure no automatic pixel conversion is inserted between perms and foobar,
301 otherwise the frames permissions might change again and the test will be
302 meaningless: add av_log(0,0,"direct=%d\n",direct) in your code to check that.
303 You can avoid the issue with something like:
305 -vf format=rgb24,perms=random,foobar
307 ...assuming your filter accepts rgb24 of course. This will make sure the
308 necessary conversion is inserted before the perms filter.
313 Adding timeline support
314 (http://ffmpeg.org/ffmpeg-filters.html#Timeline-editing) is often an easy
315 feature to add. In the most simple case, you just have to add
316 AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC to the AVFilter.flags. You can typically
317 do this when your filter does not need to save the previous context frames, or
318 basically if your filter just alters whatever goes in and doesn't need
319 previous/future information. See for instance commit 86cb986ce that adds
320 timeline support to the fieldorder filter.
322 In some cases, you might need to reset your context somehow. This is handled by
323 the AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL flag which is used if the filter
324 must not process the frames but still wants to keep track of the frames going
325 through (to keep them in cache for when it's enabled again). See for example
326 commit 69d72140a that adds timeline support to the phase filter.
331 libavfilter does not yet support frame threading, but you can add slice
332 threading to your filters.
334 Let's say the foobar filter has the following frame processing function:
339 for (y = 0; y < inlink->h; y++) {
340 for (x = 0; x < inlink->w; x++)
341 dst[x] = foobar(src[x]);
342 dst += out->linesize[0];
343 src += in ->linesize[0];
346 The first thing is to make this function work into slices. The new code will
349 for (y = slice_start; y < slice_end; y++) {
350 for (x = 0; x < inlink->w; x++)
351 dst[x] = foobar(src[x]);
352 dst += out->linesize[0];
353 src += in ->linesize[0];
356 The source and destination pointers, and slice_start/slice_end will be defined
357 according to the number of jobs. Generally, it looks like this:
359 const int slice_start = (in->height * jobnr ) / nb_jobs;
360 const int slice_end = (in->height * (jobnr+1)) / nb_jobs;
361 uint8_t *dst = out->data[0] + slice_start * out->linesize[0];
362 const uint8_t *src = in->data[0] + slice_start * in->linesize[0];
364 This new code will be isolated in a new filter_slice():
366 static int filter_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) { ... }
368 Note that we need our input and output frame to define slice_{start,end} and
369 dst/src, which are not available in that callback. They will be transmitted
370 through the opaque void *arg. You have to define a structure which contains
373 typedef struct ThreadData {
377 If you need some more information from your local context, put them here.
379 In you filter_slice function, you access it like that:
381 const ThreadData *td = arg;
383 Then in your filter_frame() callback, you need to call the threading
384 distributor with something like this:
392 ctx->internal->execute(ctx, filter_slice, &td, NULL, FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
396 return ff_filter_frame(outlink, out);
398 Last step is to add AVFILTER_FLAG_SLICE_THREADS flag to AVFilter.flags.
400 For more example of slice threading additions, you can try to run git log -p
401 --grep 'slice threading' libavfilter/
406 When your awesome filter is finished, you have a few more steps before you're
409 - write its documentation in doc/filters.texi, and test the output with make
410 doc/ffmpeg-filters.html.
411 - add a FATE test, generally by adding an entry in
412 tests/fate/filter-video.mak, add running make fate-filter-foobar GEN=1 to
414 - add an entry in the Changelog
415 - edit libavfilter/version.h and increase LIBAVFILTER_VERSION_MINOR by one
416 (and reset LIBAVFILTER_VERSION_MICRO to 100)
417 - git add ... && git commit -m "avfilter: add foobar filter." && git format-patch -1
419 When all of this is done, you can submit your patch to the ffmpeg-devel
420 mailing-list for review. If you need any help, feel free to come on our IRC
421 channel, #ffmpeg-devel on irc.freenode.net.