blob: 1fb242a3ae542ca4d20c4a02b16a0dcad692d8a1
1 | /* |
2 | * Copyright (c) 2012 Fredrik Mellbin |
3 | * Copyright (c) 2013 Clément Bœsch |
4 | * |
5 | * This file is part of FFmpeg. |
6 | * |
7 | * FFmpeg is free software; you can redistribute it and/or |
8 | * modify it under the terms of the GNU Lesser General Public |
9 | * License as published by the Free Software Foundation; either |
10 | * version 2.1 of the License, or (at your option) any later version. |
11 | * |
12 | * FFmpeg is distributed in the hope that it will be useful, |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
15 | * Lesser General Public License for more details. |
16 | * |
17 | * You should have received a copy of the GNU Lesser General Public |
18 | * License along with FFmpeg; if not, write to the Free Software |
19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
20 | */ |
21 | |
22 | #include "libavutil/opt.h" |
23 | #include "libavutil/pixdesc.h" |
24 | #include "libavutil/timestamp.h" |
25 | #include "avfilter.h" |
26 | #include "internal.h" |
27 | |
28 | #define INPUT_MAIN 0 |
29 | #define INPUT_CLEANSRC 1 |
30 | |
31 | struct qitem { |
32 | AVFrame *frame; |
33 | int64_t maxbdiff; |
34 | int64_t totdiff; |
35 | }; |
36 | |
37 | typedef struct { |
38 | const AVClass *class; |
39 | struct qitem *queue; ///< window of cycle frames and the associated data diff |
40 | int fid; ///< current frame id in the queue |
41 | int filled; ///< 1 if the queue is filled, 0 otherwise |
42 | AVFrame *last; ///< last frame from the previous queue |
43 | AVFrame **clean_src; ///< frame queue for the clean source |
44 | int got_frame[2]; ///< frame request flag for each input stream |
45 | AVRational ts_unit; ///< timestamp units for the output frames |
46 | int64_t start_pts; ///< base for output timestamps |
47 | uint32_t eof; ///< bitmask for end of stream |
48 | int hsub, vsub; ///< chroma subsampling values |
49 | int depth; |
50 | int nxblocks, nyblocks; |
51 | int bdiffsize; |
52 | int64_t *bdiffs; |
53 | |
54 | /* options */ |
55 | int cycle; |
56 | double dupthresh_flt; |
57 | double scthresh_flt; |
58 | int64_t dupthresh; |
59 | int64_t scthresh; |
60 | int blockx, blocky; |
61 | int ppsrc; |
62 | int chroma; |
63 | } DecimateContext; |
64 | |
65 | #define OFFSET(x) offsetof(DecimateContext, x) |
66 | #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM |
67 | |
68 | static const AVOption decimate_options[] = { |
69 | { "cycle", "set the number of frame from which one will be dropped", OFFSET(cycle), AV_OPT_TYPE_INT, {.i64 = 5}, 2, 25, FLAGS }, |
70 | { "dupthresh", "set duplicate threshold", OFFSET(dupthresh_flt), AV_OPT_TYPE_DOUBLE, {.dbl = 1.1}, 0, 100, FLAGS }, |
71 | { "scthresh", "set scene change threshold", OFFSET(scthresh_flt), AV_OPT_TYPE_DOUBLE, {.dbl = 15.0}, 0, 100, FLAGS }, |
72 | { "blockx", "set the size of the x-axis blocks used during metric calculations", OFFSET(blockx), AV_OPT_TYPE_INT, {.i64 = 32}, 4, 1<<9, FLAGS }, |
73 | { "blocky", "set the size of the y-axis blocks used during metric calculations", OFFSET(blocky), AV_OPT_TYPE_INT, {.i64 = 32}, 4, 1<<9, FLAGS }, |
74 | { "ppsrc", "mark main input as a pre-processed input and activate clean source input stream", OFFSET(ppsrc), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS }, |
75 | { "chroma", "set whether or not chroma is considered in the metric calculations", OFFSET(chroma), AV_OPT_TYPE_BOOL, {.i64=1}, 0, 1, FLAGS }, |
76 | { NULL } |
77 | }; |
78 | |
79 | AVFILTER_DEFINE_CLASS(decimate); |
80 | |
81 | static void calc_diffs(const DecimateContext *dm, struct qitem *q, |
82 | const AVFrame *f1, const AVFrame *f2) |
83 | { |
84 | int64_t maxdiff = -1; |
85 | int64_t *bdiffs = dm->bdiffs; |
86 | int plane, i, j; |
87 | |
88 | memset(bdiffs, 0, dm->bdiffsize * sizeof(*bdiffs)); |
89 | |
90 | for (plane = 0; plane < (dm->chroma && f1->data[2] ? 3 : 1); plane++) { |
91 | int x, y, xl; |
92 | const int linesize1 = f1->linesize[plane]; |
93 | const int linesize2 = f2->linesize[plane]; |
94 | const uint8_t *f1p = f1->data[plane]; |
95 | const uint8_t *f2p = f2->data[plane]; |
96 | int width = plane ? AV_CEIL_RSHIFT(f1->width, dm->hsub) : f1->width; |
97 | int height = plane ? AV_CEIL_RSHIFT(f1->height, dm->vsub) : f1->height; |
98 | int hblockx = dm->blockx / 2; |
99 | int hblocky = dm->blocky / 2; |
100 | |
101 | if (plane) { |
102 | hblockx >>= dm->hsub; |
103 | hblocky >>= dm->vsub; |
104 | } |
105 | |
106 | for (y = 0; y < height; y++) { |
107 | int ydest = y / hblocky; |
108 | int xdest = 0; |
109 | |
110 | #define CALC_DIFF(nbits) do { \ |
111 | for (x = 0; x < width; x += hblockx) { \ |
112 | int64_t acc = 0; \ |
113 | int m = FFMIN(width, x + hblockx); \ |
114 | for (xl = x; xl < m; xl++) \ |
115 | acc += abs(((const uint##nbits##_t *)f1p)[xl] - \ |
116 | ((const uint##nbits##_t *)f2p)[xl]); \ |
117 | bdiffs[ydest * dm->nxblocks + xdest] += acc; \ |
118 | xdest++; \ |
119 | } \ |
120 | } while (0) |
121 | if (dm->depth == 8) CALC_DIFF(8); |
122 | else CALC_DIFF(16); |
123 | |
124 | f1p += linesize1; |
125 | f2p += linesize2; |
126 | } |
127 | } |
128 | |
129 | for (i = 0; i < dm->nyblocks - 1; i++) { |
130 | for (j = 0; j < dm->nxblocks - 1; j++) { |
131 | int64_t tmp = bdiffs[ i * dm->nxblocks + j ] |
132 | + bdiffs[ i * dm->nxblocks + j + 1] |
133 | + bdiffs[(i + 1) * dm->nxblocks + j ] |
134 | + bdiffs[(i + 1) * dm->nxblocks + j + 1]; |
135 | if (tmp > maxdiff) |
136 | maxdiff = tmp; |
137 | } |
138 | } |
139 | |
140 | q->totdiff = 0; |
141 | for (i = 0; i < dm->bdiffsize; i++) |
142 | q->totdiff += bdiffs[i]; |
143 | q->maxbdiff = maxdiff; |
144 | } |
145 | |
146 | static int filter_frame(AVFilterLink *inlink, AVFrame *in) |
147 | { |
148 | int scpos = -1, duppos = -1; |
149 | int drop = INT_MIN, i, lowest = 0, ret; |
150 | AVFilterContext *ctx = inlink->dst; |
151 | AVFilterLink *outlink = ctx->outputs[0]; |
152 | DecimateContext *dm = ctx->priv; |
153 | AVFrame *prv; |
154 | |
155 | /* update frames queue(s) */ |
156 | if (FF_INLINK_IDX(inlink) == INPUT_MAIN) { |
157 | dm->queue[dm->fid].frame = in; |
158 | dm->got_frame[INPUT_MAIN] = 1; |
159 | } else { |
160 | dm->clean_src[dm->fid] = in; |
161 | dm->got_frame[INPUT_CLEANSRC] = 1; |
162 | } |
163 | if (!dm->got_frame[INPUT_MAIN] || (dm->ppsrc && !dm->got_frame[INPUT_CLEANSRC])) |
164 | return 0; |
165 | dm->got_frame[INPUT_MAIN] = dm->got_frame[INPUT_CLEANSRC] = 0; |
166 | |
167 | if (dm->ppsrc) |
168 | in = dm->clean_src[dm->fid]; |
169 | |
170 | if (in) { |
171 | /* update frame metrics */ |
172 | prv = dm->fid ? (dm->ppsrc ? dm->clean_src[dm->fid - 1] : dm->queue[dm->fid - 1].frame) : dm->last; |
173 | if (!prv) { |
174 | dm->queue[dm->fid].maxbdiff = INT64_MAX; |
175 | dm->queue[dm->fid].totdiff = INT64_MAX; |
176 | } else { |
177 | calc_diffs(dm, &dm->queue[dm->fid], prv, in); |
178 | } |
179 | if (++dm->fid != dm->cycle) |
180 | return 0; |
181 | av_frame_free(&dm->last); |
182 | dm->last = av_frame_clone(in); |
183 | dm->fid = 0; |
184 | |
185 | /* we have a complete cycle, select the frame to drop */ |
186 | lowest = 0; |
187 | for (i = 0; i < dm->cycle; i++) { |
188 | if (dm->queue[i].totdiff > dm->scthresh) |
189 | scpos = i; |
190 | if (dm->queue[i].maxbdiff < dm->queue[lowest].maxbdiff) |
191 | lowest = i; |
192 | } |
193 | if (dm->queue[lowest].maxbdiff < dm->dupthresh) |
194 | duppos = lowest; |
195 | drop = scpos >= 0 && duppos < 0 ? scpos : lowest; |
196 | } |
197 | |
198 | /* metrics debug */ |
199 | if (av_log_get_level() >= AV_LOG_DEBUG) { |
200 | av_log(ctx, AV_LOG_DEBUG, "1/%d frame drop:\n", dm->cycle); |
201 | for (i = 0; i < dm->cycle && dm->queue[i].frame; i++) { |
202 | av_log(ctx, AV_LOG_DEBUG," #%d: totdiff=%08"PRIx64" maxbdiff=%08"PRIx64"%s%s%s%s\n", |
203 | i + 1, dm->queue[i].totdiff, dm->queue[i].maxbdiff, |
204 | i == scpos ? " sc" : "", |
205 | i == duppos ? " dup" : "", |
206 | i == lowest ? " lowest" : "", |
207 | i == drop ? " [DROP]" : ""); |
208 | } |
209 | } |
210 | |
211 | /* push all frames except the drop */ |
212 | ret = 0; |
213 | for (i = 0; i < dm->cycle && dm->queue[i].frame; i++) { |
214 | if (i == drop) { |
215 | if (dm->ppsrc) |
216 | av_frame_free(&dm->clean_src[i]); |
217 | av_frame_free(&dm->queue[i].frame); |
218 | } else { |
219 | AVFrame *frame = dm->queue[i].frame; |
220 | if (frame->pts != AV_NOPTS_VALUE && dm->start_pts == AV_NOPTS_VALUE) |
221 | dm->start_pts = frame->pts; |
222 | if (dm->ppsrc) { |
223 | av_frame_free(&frame); |
224 | frame = dm->clean_src[i]; |
225 | } |
226 | frame->pts = av_rescale_q(outlink->frame_count_in, dm->ts_unit, (AVRational){1,1}) + |
227 | (dm->start_pts == AV_NOPTS_VALUE ? 0 : dm->start_pts); |
228 | ret = ff_filter_frame(outlink, frame); |
229 | if (ret < 0) |
230 | break; |
231 | } |
232 | } |
233 | |
234 | return ret; |
235 | } |
236 | |
237 | static int config_input(AVFilterLink *inlink) |
238 | { |
239 | int max_value; |
240 | AVFilterContext *ctx = inlink->dst; |
241 | DecimateContext *dm = ctx->priv; |
242 | const AVPixFmtDescriptor *pix_desc = av_pix_fmt_desc_get(inlink->format); |
243 | const int w = inlink->w; |
244 | const int h = inlink->h; |
245 | |
246 | dm->hsub = pix_desc->log2_chroma_w; |
247 | dm->vsub = pix_desc->log2_chroma_h; |
248 | dm->depth = pix_desc->comp[0].depth; |
249 | max_value = (1 << dm->depth) - 1; |
250 | dm->scthresh = (int64_t)(((int64_t)max_value * w * h * dm->scthresh_flt) / 100); |
251 | dm->dupthresh = (int64_t)(((int64_t)max_value * dm->blockx * dm->blocky * dm->dupthresh_flt) / 100); |
252 | dm->nxblocks = (w + dm->blockx/2 - 1) / (dm->blockx/2); |
253 | dm->nyblocks = (h + dm->blocky/2 - 1) / (dm->blocky/2); |
254 | dm->bdiffsize = dm->nxblocks * dm->nyblocks; |
255 | dm->bdiffs = av_malloc_array(dm->bdiffsize, sizeof(*dm->bdiffs)); |
256 | dm->queue = av_calloc(dm->cycle, sizeof(*dm->queue)); |
257 | |
258 | if (!dm->bdiffs || !dm->queue) |
259 | return AVERROR(ENOMEM); |
260 | |
261 | if (dm->ppsrc) { |
262 | dm->clean_src = av_calloc(dm->cycle, sizeof(*dm->clean_src)); |
263 | if (!dm->clean_src) |
264 | return AVERROR(ENOMEM); |
265 | } |
266 | |
267 | return 0; |
268 | } |
269 | |
270 | static av_cold int decimate_init(AVFilterContext *ctx) |
271 | { |
272 | DecimateContext *dm = ctx->priv; |
273 | AVFilterPad pad = { |
274 | .name = av_strdup("main"), |
275 | .type = AVMEDIA_TYPE_VIDEO, |
276 | .filter_frame = filter_frame, |
277 | .config_props = config_input, |
278 | }; |
279 | |
280 | if (!pad.name) |
281 | return AVERROR(ENOMEM); |
282 | ff_insert_inpad(ctx, INPUT_MAIN, &pad); |
283 | |
284 | if (dm->ppsrc) { |
285 | pad.name = av_strdup("clean_src"); |
286 | pad.config_props = NULL; |
287 | if (!pad.name) |
288 | return AVERROR(ENOMEM); |
289 | ff_insert_inpad(ctx, INPUT_CLEANSRC, &pad); |
290 | } |
291 | |
292 | if ((dm->blockx & (dm->blockx - 1)) || |
293 | (dm->blocky & (dm->blocky - 1))) { |
294 | av_log(ctx, AV_LOG_ERROR, "blockx and blocky settings must be power of two\n"); |
295 | return AVERROR(EINVAL); |
296 | } |
297 | |
298 | dm->start_pts = AV_NOPTS_VALUE; |
299 | |
300 | return 0; |
301 | } |
302 | |
303 | static av_cold void decimate_uninit(AVFilterContext *ctx) |
304 | { |
305 | int i; |
306 | DecimateContext *dm = ctx->priv; |
307 | |
308 | av_frame_free(&dm->last); |
309 | av_freep(&dm->bdiffs); |
310 | av_freep(&dm->queue); |
311 | av_freep(&dm->clean_src); |
312 | for (i = 0; i < ctx->nb_inputs; i++) |
313 | av_freep(&ctx->input_pads[i].name); |
314 | } |
315 | |
316 | static int request_inlink(AVFilterContext *ctx, int lid) |
317 | { |
318 | int ret = 0; |
319 | DecimateContext *dm = ctx->priv; |
320 | |
321 | if (!dm->got_frame[lid]) { |
322 | AVFilterLink *inlink = ctx->inputs[lid]; |
323 | ret = ff_request_frame(inlink); |
324 | if (ret == AVERROR_EOF) { // flushing |
325 | dm->eof |= 1 << lid; |
326 | ret = filter_frame(inlink, NULL); |
327 | } |
328 | } |
329 | return ret; |
330 | } |
331 | |
332 | static int request_frame(AVFilterLink *outlink) |
333 | { |
334 | int ret; |
335 | AVFilterContext *ctx = outlink->src; |
336 | DecimateContext *dm = ctx->priv; |
337 | const uint32_t eof_mask = 1<<INPUT_MAIN | dm->ppsrc<<INPUT_CLEANSRC; |
338 | |
339 | if ((dm->eof & eof_mask) == eof_mask) // flush done? |
340 | return AVERROR_EOF; |
341 | if ((ret = request_inlink(ctx, INPUT_MAIN)) < 0) |
342 | return ret; |
343 | if (dm->ppsrc && (ret = request_inlink(ctx, INPUT_CLEANSRC)) < 0) |
344 | return ret; |
345 | return 0; |
346 | } |
347 | |
348 | static int query_formats(AVFilterContext *ctx) |
349 | { |
350 | static const enum AVPixelFormat pix_fmts[] = { |
351 | #define PF_NOALPHA(suf) AV_PIX_FMT_YUV420##suf, AV_PIX_FMT_YUV422##suf, AV_PIX_FMT_YUV444##suf |
352 | #define PF_ALPHA(suf) AV_PIX_FMT_YUVA420##suf, AV_PIX_FMT_YUVA422##suf, AV_PIX_FMT_YUVA444##suf |
353 | #define PF(suf) PF_NOALPHA(suf), PF_ALPHA(suf) |
354 | PF(P), PF(P9), PF(P10), PF_NOALPHA(P12), PF_NOALPHA(P14), PF(P16), |
355 | AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P, |
356 | AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY16, |
357 | AV_PIX_FMT_NONE |
358 | }; |
359 | AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts); |
360 | if (!fmts_list) |
361 | return AVERROR(ENOMEM); |
362 | return ff_set_common_formats(ctx, fmts_list); |
363 | } |
364 | |
365 | static int config_output(AVFilterLink *outlink) |
366 | { |
367 | AVFilterContext *ctx = outlink->src; |
368 | DecimateContext *dm = ctx->priv; |
369 | const AVFilterLink *inlink = |
370 | ctx->inputs[dm->ppsrc ? INPUT_CLEANSRC : INPUT_MAIN]; |
371 | AVRational fps = inlink->frame_rate; |
372 | |
373 | if (!fps.num || !fps.den) { |
374 | av_log(ctx, AV_LOG_ERROR, "The input needs a constant frame rate; " |
375 | "current rate of %d/%d is invalid\n", fps.num, fps.den); |
376 | return AVERROR(EINVAL); |
377 | } |
378 | fps = av_mul_q(fps, (AVRational){dm->cycle - 1, dm->cycle}); |
379 | av_log(ctx, AV_LOG_VERBOSE, "FPS: %d/%d -> %d/%d\n", |
380 | inlink->frame_rate.num, inlink->frame_rate.den, fps.num, fps.den); |
381 | outlink->time_base = inlink->time_base; |
382 | outlink->frame_rate = fps; |
383 | outlink->sample_aspect_ratio = inlink->sample_aspect_ratio; |
384 | outlink->w = inlink->w; |
385 | outlink->h = inlink->h; |
386 | dm->ts_unit = av_inv_q(av_mul_q(fps, outlink->time_base)); |
387 | return 0; |
388 | } |
389 | |
390 | static const AVFilterPad decimate_outputs[] = { |
391 | { |
392 | .name = "default", |
393 | .type = AVMEDIA_TYPE_VIDEO, |
394 | .request_frame = request_frame, |
395 | .config_props = config_output, |
396 | }, |
397 | { NULL } |
398 | }; |
399 | |
400 | AVFilter ff_vf_decimate = { |
401 | .name = "decimate", |
402 | .description = NULL_IF_CONFIG_SMALL("Decimate frames (post field matching filter)."), |
403 | .init = decimate_init, |
404 | .uninit = decimate_uninit, |
405 | .priv_size = sizeof(DecimateContext), |
406 | .query_formats = query_formats, |
407 | .outputs = decimate_outputs, |
408 | .priv_class = &decimate_class, |
409 | .flags = AVFILTER_FLAG_DYNAMIC_INPUTS, |
410 | }; |
411 |