WebM Codec SDK
vpx_temporal_svc_encoder
1 /*
2  * Copyright (c) 2012 The WebM project authors. All Rights Reserved.
3  *
4  * Use of this source code is governed by a BSD-style license
5  * that can be found in the LICENSE file in the root of the source
6  * tree. An additional intellectual property rights grant can be found
7  * in the file PATENTS. All contributing project authors may
8  * be found in the AUTHORS file in the root of the source tree.
9  */
10 
11 // This is an example demonstrating how to implement a multi-layer VPx
12 // encoding scheme based on temporal scalability for video applications
13 // that benefit from a scalable bitstream.
14 
15 #include <assert.h>
16 #include <math.h>
17 #include <stdio.h>
18 #include <stdlib.h>
19 #include <string.h>
20 
21 #include "./vpx_config.h"
22 #include "../vpx_ports/vpx_timer.h"
23 #include "vpx/vp8cx.h"
24 #include "vpx/vpx_encoder.h"
25 
26 #include "../tools_common.h"
27 #include "../video_writer.h"
28 
29 static const char *exec_name;
30 
31 void usage_exit(void) {
32  exit(EXIT_FAILURE);
33 }
34 
35 // Denoiser states, for temporal denoising.
36 enum denoiserState {
37  kDenoiserOff,
38  kDenoiserOnYOnly,
39  kDenoiserOnYUV,
40  kDenoiserOnYUVAggressive,
41  kDenoiserOnAdaptive
42 };
43 
44 static int mode_to_num_layers[12] = {1, 2, 2, 3, 3, 3, 3, 5, 2, 3, 3, 3};
45 
46 // For rate control encoding stats.
47 struct RateControlMetrics {
48  // Number of input frames per layer.
49  int layer_input_frames[VPX_TS_MAX_LAYERS];
50  // Total (cumulative) number of encoded frames per layer.
51  int layer_tot_enc_frames[VPX_TS_MAX_LAYERS];
52  // Number of encoded non-key frames per layer.
53  int layer_enc_frames[VPX_TS_MAX_LAYERS];
54  // Framerate per layer layer (cumulative).
55  double layer_framerate[VPX_TS_MAX_LAYERS];
56  // Target average frame size per layer (per-frame-bandwidth per layer).
57  double layer_pfb[VPX_TS_MAX_LAYERS];
58  // Actual average frame size per layer.
59  double layer_avg_frame_size[VPX_TS_MAX_LAYERS];
60  // Average rate mismatch per layer (|target - actual| / target).
61  double layer_avg_rate_mismatch[VPX_TS_MAX_LAYERS];
62  // Actual encoding bitrate per layer (cumulative).
63  double layer_encoding_bitrate[VPX_TS_MAX_LAYERS];
64  // Average of the short-time encoder actual bitrate.
65  // TODO(marpan): Should we add these short-time stats for each layer?
66  double avg_st_encoding_bitrate;
67  // Variance of the short-time encoder actual bitrate.
68  double variance_st_encoding_bitrate;
69  // Window (number of frames) for computing short-timee encoding bitrate.
70  int window_size;
71  // Number of window measurements.
72  int window_count;
73  int layer_target_bitrate[VPX_MAX_LAYERS];
74 };
75 
76 // Note: these rate control metrics assume only 1 key frame in the
77 // sequence (i.e., first frame only). So for temporal pattern# 7
78 // (which has key frame for every frame on base layer), the metrics
79 // computation will be off/wrong.
80 // TODO(marpan): Update these metrics to account for multiple key frames
81 // in the stream.
82 static void set_rate_control_metrics(struct RateControlMetrics *rc,
83  vpx_codec_enc_cfg_t *cfg) {
84  unsigned int i = 0;
85  // Set the layer (cumulative) framerate and the target layer (non-cumulative)
86  // per-frame-bandwidth, for the rate control encoding stats below.
87  const double framerate = cfg->g_timebase.den / cfg->g_timebase.num;
88  rc->layer_framerate[0] = framerate / cfg->ts_rate_decimator[0];
89  rc->layer_pfb[0] = 1000.0 * rc->layer_target_bitrate[0] /
90  rc->layer_framerate[0];
91  for (i = 0; i < cfg->ts_number_layers; ++i) {
92  if (i > 0) {
93  rc->layer_framerate[i] = framerate / cfg->ts_rate_decimator[i];
94  rc->layer_pfb[i] = 1000.0 *
95  (rc->layer_target_bitrate[i] - rc->layer_target_bitrate[i - 1]) /
96  (rc->layer_framerate[i] - rc->layer_framerate[i - 1]);
97  }
98  rc->layer_input_frames[i] = 0;
99  rc->layer_enc_frames[i] = 0;
100  rc->layer_tot_enc_frames[i] = 0;
101  rc->layer_encoding_bitrate[i] = 0.0;
102  rc->layer_avg_frame_size[i] = 0.0;
103  rc->layer_avg_rate_mismatch[i] = 0.0;
104  }
105  rc->window_count = 0;
106  rc->window_size = 15;
107  rc->avg_st_encoding_bitrate = 0.0;
108  rc->variance_st_encoding_bitrate = 0.0;
109 }
110 
111 static void printout_rate_control_summary(struct RateControlMetrics *rc,
112  vpx_codec_enc_cfg_t *cfg,
113  int frame_cnt) {
114  unsigned int i = 0;
115  int tot_num_frames = 0;
116  double perc_fluctuation = 0.0;
117  printf("Total number of processed frames: %d\n\n", frame_cnt -1);
118  printf("Rate control layer stats for %d layer(s):\n\n",
119  cfg->ts_number_layers);
120  for (i = 0; i < cfg->ts_number_layers; ++i) {
121  const int num_dropped = (i > 0) ?
122  (rc->layer_input_frames[i] - rc->layer_enc_frames[i]) :
123  (rc->layer_input_frames[i] - rc->layer_enc_frames[i] - 1);
124  tot_num_frames += rc->layer_input_frames[i];
125  rc->layer_encoding_bitrate[i] = 0.001 * rc->layer_framerate[i] *
126  rc->layer_encoding_bitrate[i] / tot_num_frames;
127  rc->layer_avg_frame_size[i] = rc->layer_avg_frame_size[i] /
128  rc->layer_enc_frames[i];
129  rc->layer_avg_rate_mismatch[i] = 100.0 * rc->layer_avg_rate_mismatch[i] /
130  rc->layer_enc_frames[i];
131  printf("For layer#: %d \n", i);
132  printf("Bitrate (target vs actual): %d %f \n", rc->layer_target_bitrate[i],
133  rc->layer_encoding_bitrate[i]);
134  printf("Average frame size (target vs actual): %f %f \n", rc->layer_pfb[i],
135  rc->layer_avg_frame_size[i]);
136  printf("Average rate_mismatch: %f \n", rc->layer_avg_rate_mismatch[i]);
137  printf("Number of input frames, encoded (non-key) frames, "
138  "and perc dropped frames: %d %d %f \n", rc->layer_input_frames[i],
139  rc->layer_enc_frames[i],
140  100.0 * num_dropped / rc->layer_input_frames[i]);
141  printf("\n");
142  }
143  rc->avg_st_encoding_bitrate = rc->avg_st_encoding_bitrate / rc->window_count;
144  rc->variance_st_encoding_bitrate =
145  rc->variance_st_encoding_bitrate / rc->window_count -
146  (rc->avg_st_encoding_bitrate * rc->avg_st_encoding_bitrate);
147  perc_fluctuation = 100.0 * sqrt(rc->variance_st_encoding_bitrate) /
148  rc->avg_st_encoding_bitrate;
149  printf("Short-time stats, for window of %d frames: \n",rc->window_size);
150  printf("Average, rms-variance, and percent-fluct: %f %f %f \n",
151  rc->avg_st_encoding_bitrate,
152  sqrt(rc->variance_st_encoding_bitrate),
153  perc_fluctuation);
154  if ((frame_cnt - 1) != tot_num_frames)
155  die("Error: Number of input frames not equal to output! \n");
156 }
157 
158 // Temporal scaling parameters:
159 // NOTE: The 3 prediction frames cannot be used interchangeably due to
160 // differences in the way they are handled throughout the code. The
161 // frames should be allocated to layers in the order LAST, GF, ARF.
162 // Other combinations work, but may produce slightly inferior results.
163 static void set_temporal_layer_pattern(int layering_mode,
164  vpx_codec_enc_cfg_t *cfg,
165  int *layer_flags,
166  int *flag_periodicity) {
167  switch (layering_mode) {
168  case 0: {
169  // 1-layer.
170  int ids[1] = {0};
171  cfg->ts_periodicity = 1;
172  *flag_periodicity = 1;
173  cfg->ts_number_layers = 1;
174  cfg->ts_rate_decimator[0] = 1;
175  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
176  // Update L only.
177  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_UPD_GF |
179  break;
180  }
181  case 1: {
182  // 2-layers, 2-frame period.
183  int ids[2] = {0, 1};
184  cfg->ts_periodicity = 2;
185  *flag_periodicity = 2;
186  cfg->ts_number_layers = 2;
187  cfg->ts_rate_decimator[0] = 2;
188  cfg->ts_rate_decimator[1] = 1;
189  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
190 #if 1
191  // 0=L, 1=GF, Intra-layer prediction enabled.
192  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_UPD_GF |
194  layer_flags[1] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST |
196 #else
197  // 0=L, 1=GF, Intra-layer prediction disabled.
198  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_UPD_GF |
200  layer_flags[1] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST |
202 #endif
203  break;
204  }
205  case 2: {
206  // 2-layers, 3-frame period.
207  int ids[3] = {0, 1, 1};
208  cfg->ts_periodicity = 3;
209  *flag_periodicity = 3;
210  cfg->ts_number_layers = 2;
211  cfg->ts_rate_decimator[0] = 3;
212  cfg->ts_rate_decimator[1] = 1;
213  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
214  // 0=L, 1=GF, Intra-layer prediction enabled.
215  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_REF_GF |
217  layer_flags[1] =
218  layer_flags[2] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
220  break;
221  }
222  case 3: {
223  // 3-layers, 6-frame period.
224  int ids[6] = {0, 2, 2, 1, 2, 2};
225  cfg->ts_periodicity = 6;
226  *flag_periodicity = 6;
227  cfg->ts_number_layers = 3;
228  cfg->ts_rate_decimator[0] = 6;
229  cfg->ts_rate_decimator[1] = 3;
230  cfg->ts_rate_decimator[2] = 1;
231  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
232  // 0=L, 1=GF, 2=ARF, Intra-layer prediction enabled.
233  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_REF_GF |
235  layer_flags[3] = VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_ARF |
237  layer_flags[1] =
238  layer_flags[2] =
239  layer_flags[4] =
240  layer_flags[5] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_LAST;
241  break;
242  }
243  case 4: {
244  // 3-layers, 4-frame period.
245  int ids[4] = {0, 2, 1, 2};
246  cfg->ts_periodicity = 4;
247  *flag_periodicity = 4;
248  cfg->ts_number_layers = 3;
249  cfg->ts_rate_decimator[0] = 4;
250  cfg->ts_rate_decimator[1] = 2;
251  cfg->ts_rate_decimator[2] = 1;
252  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
253  // 0=L, 1=GF, 2=ARF, Intra-layer prediction disabled.
254  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_REF_GF |
256  layer_flags[2] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
258  layer_flags[1] =
259  layer_flags[3] = VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_LAST |
261  break;
262  }
263  case 5: {
264  // 3-layers, 4-frame period.
265  int ids[4] = {0, 2, 1, 2};
266  cfg->ts_periodicity = 4;
267  *flag_periodicity = 4;
268  cfg->ts_number_layers = 3;
269  cfg->ts_rate_decimator[0] = 4;
270  cfg->ts_rate_decimator[1] = 2;
271  cfg->ts_rate_decimator[2] = 1;
272  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
273  // 0=L, 1=GF, 2=ARF, Intra-layer prediction enabled in layer 1, disabled
274  // in layer 2.
275  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_REF_GF |
277  layer_flags[2] = VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_LAST |
279  layer_flags[1] =
280  layer_flags[3] = VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_LAST |
282  break;
283  }
284  case 6: {
285  // 3-layers, 4-frame period.
286  int ids[4] = {0, 2, 1, 2};
287  cfg->ts_periodicity = 4;
288  *flag_periodicity = 4;
289  cfg->ts_number_layers = 3;
290  cfg->ts_rate_decimator[0] = 4;
291  cfg->ts_rate_decimator[1] = 2;
292  cfg->ts_rate_decimator[2] = 1;
293  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
294  // 0=L, 1=GF, 2=ARF, Intra-layer prediction enabled.
295  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_REF_GF |
297  layer_flags[2] = VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_LAST |
299  layer_flags[1] =
300  layer_flags[3] = VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF;
301  break;
302  }
303  case 7: {
304  // NOTE: Probably of academic interest only.
305  // 5-layers, 16-frame period.
306  int ids[16] = {0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4};
307  cfg->ts_periodicity = 16;
308  *flag_periodicity = 16;
309  cfg->ts_number_layers = 5;
310  cfg->ts_rate_decimator[0] = 16;
311  cfg->ts_rate_decimator[1] = 8;
312  cfg->ts_rate_decimator[2] = 4;
313  cfg->ts_rate_decimator[3] = 2;
314  cfg->ts_rate_decimator[4] = 1;
315  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
316  layer_flags[0] = VPX_EFLAG_FORCE_KF;
317  layer_flags[1] =
318  layer_flags[3] =
319  layer_flags[5] =
320  layer_flags[7] =
321  layer_flags[9] =
322  layer_flags[11] =
323  layer_flags[13] =
324  layer_flags[15] = VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF |
326  layer_flags[2] =
327  layer_flags[6] =
328  layer_flags[10] =
329  layer_flags[14] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_GF;
330  layer_flags[4] =
331  layer_flags[12] = VP8_EFLAG_NO_REF_LAST | VP8_EFLAG_NO_UPD_ARF;
332  layer_flags[8] = VP8_EFLAG_NO_REF_LAST | VP8_EFLAG_NO_REF_GF;
333  break;
334  }
335  case 8: {
336  // 2-layers, with sync point at first frame of layer 1.
337  int ids[2] = {0, 1};
338  cfg->ts_periodicity = 2;
339  *flag_periodicity = 8;
340  cfg->ts_number_layers = 2;
341  cfg->ts_rate_decimator[0] = 2;
342  cfg->ts_rate_decimator[1] = 1;
343  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
344  // 0=L, 1=GF.
345  // ARF is used as predictor for all frames, and is only updated on
346  // key frame. Sync point every 8 frames.
347 
348  // Layer 0: predict from L and ARF, update L and G.
349  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_REF_GF |
351  // Layer 1: sync point: predict from L and ARF, and update G.
352  layer_flags[1] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_UPD_LAST |
354  // Layer 0, predict from L and ARF, update L.
355  layer_flags[2] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_UPD_GF |
357  // Layer 1: predict from L, G and ARF, and update G.
358  layer_flags[3] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST |
360  // Layer 0.
361  layer_flags[4] = layer_flags[2];
362  // Layer 1.
363  layer_flags[5] = layer_flags[3];
364  // Layer 0.
365  layer_flags[6] = layer_flags[4];
366  // Layer 1.
367  layer_flags[7] = layer_flags[5];
368  break;
369  }
370  case 9: {
371  // 3-layers: Sync points for layer 1 and 2 every 8 frames.
372  int ids[4] = {0, 2, 1, 2};
373  cfg->ts_periodicity = 4;
374  *flag_periodicity = 8;
375  cfg->ts_number_layers = 3;
376  cfg->ts_rate_decimator[0] = 4;
377  cfg->ts_rate_decimator[1] = 2;
378  cfg->ts_rate_decimator[2] = 1;
379  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
380  // 0=L, 1=GF, 2=ARF.
381  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_REF_GF |
383  layer_flags[1] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
385  layer_flags[2] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
387  layer_flags[3] =
388  layer_flags[5] = VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF;
389  layer_flags[4] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
391  layer_flags[6] = VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_LAST |
393  layer_flags[7] = VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF |
395  break;
396  }
397  case 10: {
398  // 3-layers structure where ARF is used as predictor for all frames,
399  // and is only updated on key frame.
400  // Sync points for layer 1 and 2 every 8 frames.
401 
402  int ids[4] = {0, 2, 1, 2};
403  cfg->ts_periodicity = 4;
404  *flag_periodicity = 8;
405  cfg->ts_number_layers = 3;
406  cfg->ts_rate_decimator[0] = 4;
407  cfg->ts_rate_decimator[1] = 2;
408  cfg->ts_rate_decimator[2] = 1;
409  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
410  // 0=L, 1=GF, 2=ARF.
411  // Layer 0: predict from L and ARF; update L and G.
412  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_UPD_ARF |
414  // Layer 2: sync point: predict from L and ARF; update none.
415  layer_flags[1] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_UPD_GF |
418  // Layer 1: sync point: predict from L and ARF; update G.
419  layer_flags[2] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_UPD_ARF |
421  // Layer 2: predict from L, G, ARF; update none.
422  layer_flags[3] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
424  // Layer 0: predict from L and ARF; update L.
425  layer_flags[4] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
427  // Layer 2: predict from L, G, ARF; update none.
428  layer_flags[5] = layer_flags[3];
429  // Layer 1: predict from L, G, ARF; update G.
430  layer_flags[6] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST;
431  // Layer 2: predict from L, G, ARF; update none.
432  layer_flags[7] = layer_flags[3];
433  break;
434  }
435  case 11:
436  default: {
437  // 3-layers structure as in case 10, but no sync/refresh points for
438  // layer 1 and 2.
439  int ids[4] = {0, 2, 1, 2};
440  cfg->ts_periodicity = 4;
441  *flag_periodicity = 8;
442  cfg->ts_number_layers = 3;
443  cfg->ts_rate_decimator[0] = 4;
444  cfg->ts_rate_decimator[1] = 2;
445  cfg->ts_rate_decimator[2] = 1;
446  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
447  // 0=L, 1=GF, 2=ARF.
448  // Layer 0: predict from L and ARF; update L.
449  layer_flags[0] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
451  layer_flags[4] = layer_flags[0];
452  // Layer 1: predict from L, G, ARF; update G.
453  layer_flags[2] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST;
454  layer_flags[6] = layer_flags[2];
455  // Layer 2: predict from L, G, ARF; update none.
456  layer_flags[1] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
458  layer_flags[3] = layer_flags[1];
459  layer_flags[5] = layer_flags[1];
460  layer_flags[7] = layer_flags[1];
461  break;
462  }
463  }
464 }
465 
466 int main(int argc, char **argv) {
467  VpxVideoWriter *outfile[VPX_TS_MAX_LAYERS] = {NULL};
468  vpx_codec_ctx_t codec;
470  int frame_cnt = 0;
471  vpx_image_t raw;
472  vpx_codec_err_t res;
473  unsigned int width;
474  unsigned int height;
475  int speed;
476  int frame_avail;
477  int got_data;
478  int flags = 0;
479  unsigned int i;
480  int pts = 0; // PTS starts at 0.
481  int frame_duration = 1; // 1 timebase tick per frame.
482  int layering_mode = 0;
483  int layer_flags[VPX_TS_MAX_PERIODICITY] = {0};
484  int flag_periodicity = 1;
485 #if VPX_ENCODER_ABI_VERSION > (4 + VPX_CODEC_ABI_VERSION)
486  vpx_svc_layer_id_t layer_id = {0, 0};
487 #else
488  vpx_svc_layer_id_t layer_id = {0};
489 #endif
490  const VpxInterface *encoder = NULL;
491  FILE *infile = NULL;
492  struct RateControlMetrics rc;
493  int64_t cx_time = 0;
494  const int min_args_base = 11;
495 #if CONFIG_VP9_HIGHBITDEPTH
496  vpx_bit_depth_t bit_depth = VPX_BITS_8;
497  int input_bit_depth = 8;
498  const int min_args = min_args_base + 1;
499 #else
500  const int min_args = min_args_base;
501 #endif // CONFIG_VP9_HIGHBITDEPTH
502  double sum_bitrate = 0.0;
503  double sum_bitrate2 = 0.0;
504  double framerate = 30.0;
505 
506  exec_name = argv[0];
507  // Check usage and arguments.
508  if (argc < min_args) {
509 #if CONFIG_VP9_HIGHBITDEPTH
510  die("Usage: %s <infile> <outfile> <codec_type(vp8/vp9)> <width> <height> "
511  "<rate_num> <rate_den> <speed> <frame_drop_threshold> <mode> "
512  "<Rate_0> ... <Rate_nlayers-1> <bit-depth> \n", argv[0]);
513 #else
514  die("Usage: %s <infile> <outfile> <codec_type(vp8/vp9)> <width> <height> "
515  "<rate_num> <rate_den> <speed> <frame_drop_threshold> <mode> "
516  "<Rate_0> ... <Rate_nlayers-1> \n", argv[0]);
517 #endif // CONFIG_VP9_HIGHBITDEPTH
518  }
519 
520  encoder = get_vpx_encoder_by_name(argv[3]);
521  if (!encoder)
522  die("Unsupported codec.");
523 
524  printf("Using %s\n", vpx_codec_iface_name(encoder->codec_interface()));
525 
526  width = strtol(argv[4], NULL, 0);
527  height = strtol(argv[5], NULL, 0);
528  if (width < 16 || width % 2 || height < 16 || height % 2) {
529  die("Invalid resolution: %d x %d", width, height);
530  }
531 
532  layering_mode = strtol(argv[10], NULL, 0);
533  if (layering_mode < 0 || layering_mode > 12) {
534  die("Invalid layering mode (0..12) %s", argv[10]);
535  }
536 
537  if (argc != min_args + mode_to_num_layers[layering_mode]) {
538  die("Invalid number of arguments");
539  }
540 
541 #if CONFIG_VP9_HIGHBITDEPTH
542  switch (strtol(argv[argc-1], NULL, 0)) {
543  case 8:
544  bit_depth = VPX_BITS_8;
545  input_bit_depth = 8;
546  break;
547  case 10:
548  bit_depth = VPX_BITS_10;
549  input_bit_depth = 10;
550  break;
551  case 12:
552  bit_depth = VPX_BITS_12;
553  input_bit_depth = 12;
554  break;
555  default:
556  die("Invalid bit depth (8, 10, 12) %s", argv[argc-1]);
557  }
558  if (!vpx_img_alloc(&raw,
559  bit_depth == VPX_BITS_8 ? VPX_IMG_FMT_I420 :
560  VPX_IMG_FMT_I42016,
561  width, height, 32)) {
562  die("Failed to allocate image", width, height);
563  }
564 #else
565  if (!vpx_img_alloc(&raw, VPX_IMG_FMT_I420, width, height, 32)) {
566  die("Failed to allocate image", width, height);
567  }
568 #endif // CONFIG_VP9_HIGHBITDEPTH
569 
570  // Populate encoder configuration.
571  res = vpx_codec_enc_config_default(encoder->codec_interface(), &cfg, 0);
572  if (res) {
573  printf("Failed to get config: %s\n", vpx_codec_err_to_string(res));
574  return EXIT_FAILURE;
575  }
576 
577  // Update the default configuration with our settings.
578  cfg.g_w = width;
579  cfg.g_h = height;
580 
581 #if CONFIG_VP9_HIGHBITDEPTH
582  if (bit_depth != VPX_BITS_8) {
583  cfg.g_bit_depth = bit_depth;
584  cfg.g_input_bit_depth = input_bit_depth;
585  cfg.g_profile = 2;
586  }
587 #endif // CONFIG_VP9_HIGHBITDEPTH
588 
589  // Timebase format e.g. 30fps: numerator=1, demoninator = 30.
590  cfg.g_timebase.num = strtol(argv[6], NULL, 0);
591  cfg.g_timebase.den = strtol(argv[7], NULL, 0);
592 
593  speed = strtol(argv[8], NULL, 0);
594  if (speed < 0) {
595  die("Invalid speed setting: must be positive");
596  }
597 
598  for (i = min_args_base;
599  (int)i < min_args_base + mode_to_num_layers[layering_mode];
600  ++i) {
601  rc.layer_target_bitrate[i - 11] = strtol(argv[i], NULL, 0);
602  if (strncmp(encoder->name, "vp8", 3) == 0)
603  cfg.ts_target_bitrate[i - 11] = rc.layer_target_bitrate[i - 11];
604  else if (strncmp(encoder->name, "vp9", 3) == 0)
605  cfg.layer_target_bitrate[i - 11] = rc.layer_target_bitrate[i - 11];
606  }
607 
608  // Real time parameters.
609  cfg.rc_dropframe_thresh = strtol(argv[9], NULL, 0);
610  cfg.rc_end_usage = VPX_CBR;
611  cfg.rc_min_quantizer = 2;
612  cfg.rc_max_quantizer = 56;
613  if (strncmp(encoder->name, "vp9", 3) == 0)
614  cfg.rc_max_quantizer = 52;
615  cfg.rc_undershoot_pct = 50;
616  cfg.rc_overshoot_pct = 50;
617  cfg.rc_buf_initial_sz = 500;
618  cfg.rc_buf_optimal_sz = 600;
619  cfg.rc_buf_sz = 1000;
620 
621  // Disable dynamic resizing by default.
622  cfg.rc_resize_allowed = 0;
623 
624  // Use 1 thread as default.
625  cfg.g_threads = 1;
626 
627  // Enable error resilient mode.
628  cfg.g_error_resilient = 1;
629  cfg.g_lag_in_frames = 0;
630  cfg.kf_mode = VPX_KF_AUTO;
631 
632  // Disable automatic keyframe placement.
633  cfg.kf_min_dist = cfg.kf_max_dist = 3000;
634 
636 
637  set_temporal_layer_pattern(layering_mode,
638  &cfg,
639  layer_flags,
640  &flag_periodicity);
641 
642  set_rate_control_metrics(&rc, &cfg);
643 
644  // Target bandwidth for the whole stream.
645  // Set to layer_target_bitrate for highest layer (total bitrate).
646  cfg.rc_target_bitrate = rc.layer_target_bitrate[cfg.ts_number_layers - 1];
647 
648  // Open input file.
649  if (!(infile = fopen(argv[1], "rb"))) {
650  die("Failed to open %s for reading", argv[1]);
651  }
652 
653  framerate = cfg.g_timebase.den / cfg.g_timebase.num;
654  // Open an output file for each stream.
655  for (i = 0; i < cfg.ts_number_layers; ++i) {
656  char file_name[PATH_MAX];
657  VpxVideoInfo info;
658  info.codec_fourcc = encoder->fourcc;
659  info.frame_width = cfg.g_w;
660  info.frame_height = cfg.g_h;
661  info.time_base.numerator = cfg.g_timebase.num;
662  info.time_base.denominator = cfg.g_timebase.den;
663 
664  snprintf(file_name, sizeof(file_name), "%s_%d.ivf", argv[2], i);
665  outfile[i] = vpx_video_writer_open(file_name, kContainerIVF, &info);
666  if (!outfile[i])
667  die("Failed to open %s for writing", file_name);
668 
669  assert(outfile[i] != NULL);
670  }
671  // No spatial layers in this encoder.
672  cfg.ss_number_layers = 1;
673 
674  // Initialize codec.
675 #if CONFIG_VP9_HIGHBITDEPTH
676  if (vpx_codec_enc_init(
677  &codec, encoder->codec_interface(), &cfg,
678  bit_depth == VPX_BITS_8 ? 0 : VPX_CODEC_USE_HIGHBITDEPTH))
679 #else
680  if (vpx_codec_enc_init(&codec, encoder->codec_interface(), &cfg, 0))
681 #endif // CONFIG_VP9_HIGHBITDEPTH
682  die_codec(&codec, "Failed to initialize encoder");
683 
684  if (strncmp(encoder->name, "vp8", 3) == 0) {
685  vpx_codec_control(&codec, VP8E_SET_CPUUSED, -speed);
686  vpx_codec_control(&codec, VP8E_SET_NOISE_SENSITIVITY, kDenoiserOff);
688  } else if (strncmp(encoder->name, "vp9", 3) == 0) {
689  vpx_svc_extra_cfg_t svc_params;
690  vpx_codec_control(&codec, VP8E_SET_CPUUSED, speed);
697  if (vpx_codec_control(&codec, VP9E_SET_SVC, layering_mode > 0 ? 1: 0))
698  die_codec(&codec, "Failed to set SVC");
699  for (i = 0; i < cfg.ts_number_layers; ++i) {
700  svc_params.max_quantizers[i] = cfg.rc_max_quantizer;
701  svc_params.min_quantizers[i] = cfg.rc_min_quantizer;
702  }
703  svc_params.scaling_factor_num[0] = cfg.g_h;
704  svc_params.scaling_factor_den[0] = cfg.g_h;
705  vpx_codec_control(&codec, VP9E_SET_SVC_PARAMETERS, &svc_params);
706  }
707  if (strncmp(encoder->name, "vp8", 3) == 0) {
709  }
711  // This controls the maximum target size of the key frame.
712  // For generating smaller key frames, use a smaller max_intra_size_pct
713  // value, like 100 or 200.
714  {
715  const int max_intra_size_pct = 900;
717  max_intra_size_pct);
718  }
719 
720  frame_avail = 1;
721  while (frame_avail || got_data) {
722  struct vpx_usec_timer timer;
723  vpx_codec_iter_t iter = NULL;
724  const vpx_codec_cx_pkt_t *pkt;
725 #if VPX_ENCODER_ABI_VERSION > (4 + VPX_CODEC_ABI_VERSION)
726  // Update the temporal layer_id. No spatial layers in this test.
727  layer_id.spatial_layer_id = 0;
728 #endif
729  layer_id.temporal_layer_id =
730  cfg.ts_layer_id[frame_cnt % cfg.ts_periodicity];
731  if (strncmp(encoder->name, "vp9", 3) == 0) {
732  vpx_codec_control(&codec, VP9E_SET_SVC_LAYER_ID, &layer_id);
733  } else if (strncmp(encoder->name, "vp8", 3) == 0) {
735  layer_id.temporal_layer_id);
736  }
737  flags = layer_flags[frame_cnt % flag_periodicity];
738  if (layering_mode == 0)
739  flags = 0;
740  frame_avail = vpx_img_read(&raw, infile);
741  if (frame_avail)
742  ++rc.layer_input_frames[layer_id.temporal_layer_id];
743  vpx_usec_timer_start(&timer);
744  if (vpx_codec_encode(&codec, frame_avail? &raw : NULL, pts, 1, flags,
745  VPX_DL_REALTIME)) {
746  die_codec(&codec, "Failed to encode frame");
747  }
748  vpx_usec_timer_mark(&timer);
749  cx_time += vpx_usec_timer_elapsed(&timer);
750  // Reset KF flag.
751  if (layering_mode != 7) {
752  layer_flags[0] &= ~VPX_EFLAG_FORCE_KF;
753  }
754  got_data = 0;
755  while ( (pkt = vpx_codec_get_cx_data(&codec, &iter)) ) {
756  got_data = 1;
757  switch (pkt->kind) {
759  for (i = cfg.ts_layer_id[frame_cnt % cfg.ts_periodicity];
760  i < cfg.ts_number_layers; ++i) {
761  vpx_video_writer_write_frame(outfile[i], pkt->data.frame.buf,
762  pkt->data.frame.sz, pts);
763  ++rc.layer_tot_enc_frames[i];
764  rc.layer_encoding_bitrate[i] += 8.0 * pkt->data.frame.sz;
765  // Keep count of rate control stats per layer (for non-key frames).
766  if (i == cfg.ts_layer_id[frame_cnt % cfg.ts_periodicity] &&
767  !(pkt->data.frame.flags & VPX_FRAME_IS_KEY)) {
768  rc.layer_avg_frame_size[i] += 8.0 * pkt->data.frame.sz;
769  rc.layer_avg_rate_mismatch[i] +=
770  fabs(8.0 * pkt->data.frame.sz - rc.layer_pfb[i]) /
771  rc.layer_pfb[i];
772  ++rc.layer_enc_frames[i];
773  }
774  }
775  // Update for short-time encoding bitrate states, for moving window
776  // of size rc->window, shifted by rc->window / 2.
777  // Ignore first window segment, due to key frame.
778  if (frame_cnt > rc.window_size) {
779  sum_bitrate += 0.001 * 8.0 * pkt->data.frame.sz * framerate;
780  if (frame_cnt % rc.window_size == 0) {
781  rc.window_count += 1;
782  rc.avg_st_encoding_bitrate += sum_bitrate / rc.window_size;
783  rc.variance_st_encoding_bitrate +=
784  (sum_bitrate / rc.window_size) *
785  (sum_bitrate / rc.window_size);
786  sum_bitrate = 0.0;
787  }
788  }
789  // Second shifted window.
790  if (frame_cnt > rc.window_size + rc.window_size / 2) {
791  sum_bitrate2 += 0.001 * 8.0 * pkt->data.frame.sz * framerate;
792  if (frame_cnt > 2 * rc.window_size &&
793  frame_cnt % rc.window_size == 0) {
794  rc.window_count += 1;
795  rc.avg_st_encoding_bitrate += sum_bitrate2 / rc.window_size;
796  rc.variance_st_encoding_bitrate +=
797  (sum_bitrate2 / rc.window_size) *
798  (sum_bitrate2 / rc.window_size);
799  sum_bitrate2 = 0.0;
800  }
801  }
802  break;
803  default:
804  break;
805  }
806  }
807  ++frame_cnt;
808  pts += frame_duration;
809  }
810  fclose(infile);
811  printout_rate_control_summary(&rc, &cfg, frame_cnt);
812  printf("\n");
813  printf("Frame cnt and encoding time/FPS stats for encoding: %d %f %f \n",
814  frame_cnt,
815  1000 * (float)cx_time / (double)(frame_cnt * 1000000),
816  1000000 * (double)frame_cnt / (double)cx_time);
817 
818  if (vpx_codec_destroy(&codec))
819  die_codec(&codec, "Failed to destroy codec");
820 
821  // Try to rewrite the output file headers with the actual frame count.
822  for (i = 0; i < cfg.ts_number_layers; ++i)
823  vpx_video_writer_close(outfile[i]);
824 
825  vpx_img_free(&raw);
826  return EXIT_SUCCESS;
827 }
unsigned int rc_buf_initial_sz
Decoder Buffer Initial Size.
Definition: vpx_encoder.h:610
int min_quantizers[12]
Definition: vpx_encoder.h:773
unsigned int ts_number_layers
Number of temporal coding layers.
Definition: vpx_encoder.h:715
Codec control function to set encoder internal speed settings.
Definition: vp8cx.h:173
#define VPX_MAX_LAYERS
Definition: vpx_encoder.h:46
#define VP8_EFLAG_NO_REF_LAST
Don't reference the last frame.
Definition: vp8cx.h:67
#define VP8_EFLAG_NO_UPD_GF
Don't update the golden frame.
Definition: vp8cx.h:101
Image Descriptor.
Definition: vpx_image.h:88
Describes the encoder algorithm interface to applications.
const char * vpx_codec_iface_name(vpx_codec_iface_t *iface)
Return the name for a given interface.
const char * vpx_codec_err_to_string(vpx_codec_err_t err)
Convert error number to printable string.
#define VPX_TS_MAX_LAYERS
Definition: vpx_encoder.h:40
Codec control function to set content type.
Definition: vp8cx.h:467
struct vpx_rational g_timebase
Stream timebase units.
Definition: vpx_encoder.h:397
Definition: vpx_encoder.h:276
Codec control function to set noise sensitivity.
Definition: vp8cx.h:432
unsigned int layer_target_bitrate[12]
Target bitrate for each spatial/temporal layer.
Definition: vpx_encoder.h:755
unsigned int rc_buf_sz
Decoder Buffer Size.
Definition: vpx_encoder.h:600
#define VP8_EFLAG_NO_REF_GF
Don't reference the golden frame.
Definition: vp8cx.h:76
unsigned int g_input_bit_depth
Bit-depth of the input frames.
Definition: vpx_encoder.h:383
enum vpx_kf_mode kf_mode
Keyframe placement mode.
Definition: vpx_encoder.h:665
int den
Definition: vpx_encoder.h:261
vpx_codec_err_t vpx_codec_encode(vpx_codec_ctx_t *ctx, const vpx_image_t *img, vpx_codec_pts_t pts, unsigned long duration, vpx_enc_frame_flags_t flags, unsigned long deadline)
Encode a frame.
unsigned int rc_max_quantizer
Maximum (Worst Quality) Quantizer.
Definition: vpx_encoder.h:552
unsigned int rc_min_quantizer
Minimum (Best Quality) Quantizer.
Definition: vpx_encoder.h:541
unsigned int kf_max_dist
Keyframe maximum interval.
Definition: vpx_encoder.h:685
unsigned int g_lag_in_frames
Allow lagged encoding.
Definition: vpx_encoder.h:429
Encoder configuration structure.
Definition: vpx_encoder.h:314
Definition: vpx_encoder.h:292
int spatial_layer_id
Definition: vp8cx.h:679
Codec control function to set Max data rate for Intra frames.
Definition: vp8cx.h:269
#define VPX_CODEC_USE_HIGHBITDEPTH
Definition: vpx_encoder.h:101
Encoder output packet.
Definition: vpx_encoder.h:195
unsigned int rc_overshoot_pct
Rate control adaptation overshoot control.
Definition: vpx_encoder.h:583
Codec control function to set parameters for SVC.
Definition: vp8cx.h:449
unsigned int ts_rate_decimator[5]
Frame rate decimation factor for each temporal layer.
Definition: vpx_encoder.h:729
unsigned int rc_buf_optimal_sz
Decoder Buffer Optimal Size.
Definition: vpx_encoder.h:620
unsigned int kf_min_dist
Keyframe minimum interval.
Definition: vpx_encoder.h:675
unsigned int g_profile
Bitstream profile to use.
Definition: vpx_encoder.h:346
Codec control function to set number of tile columns.
Definition: vp8cx.h:362
unsigned int ts_layer_id[16]
Template defining the membership of frames to temporal layers.
Definition: vpx_encoder.h:747
struct vpx_codec_cx_pkt::@1::@2 frame
vpx_image_t * vpx_img_alloc(vpx_image_t *img, vpx_img_fmt_t fmt, unsigned int d_w, unsigned int d_h, unsigned int align)
Open a descriptor, allocating storage for the underlying image.
Definition: vpx_image.h:56
int scaling_factor_num[12]
Definition: vpx_encoder.h:774
unsigned int g_w
Width of the frame.
Definition: vpx_encoder.h:357
unsigned int ts_target_bitrate[5]
Target bitrate for each temporal layer.
Definition: vpx_encoder.h:722
enum vpx_bit_depth vpx_bit_depth_t
Bit depth for codecThis enumeration determines the bit depth of the codec.
unsigned int rc_undershoot_pct
Rate control adaptation undershoot control.
Definition: vpx_encoder.h:570
Codec control function to set adaptive quantization mode.
Definition: vp8cx.h:409
unsigned int g_h
Height of the frame.
Definition: vpx_encoder.h:367
enum vpx_codec_cx_pkt_kind kind
Definition: vpx_encoder.h:196
unsigned int rc_dropframe_thresh
Temporal resampling configuration, if supported by the codec.
Definition: vpx_encoder.h:452
vp9 svc layer parameters
Definition: vp8cx.h:678
Codec control function to set the temporal layer id.
Definition: vp8cx.h:316
#define VP8_EFLAG_NO_UPD_LAST
Don't update the last frame.
Definition: vp8cx.h:93
void vpx_img_free(vpx_image_t *img)
Close an image descriptor.
Codec control function to set the number of token partitions.
Definition: vp8cx.h:206
unsigned int rc_target_bitrate
Target data rate.
Definition: vpx_encoder.h:525
#define VPX_DL_REALTIME
Definition: vpx_encoder.h:911
int num
Definition: vpx_encoder.h:260
control function to set noise sensitivity
Definition: vp8cx.h:188
Definition: vpx_codec.h:222
unsigned int g_threads
Maximum number of threads to use.
Definition: vpx_encoder.h:335
unsigned int ss_number_layers
Number of spatial coding layers.
Definition: vpx_encoder.h:695
vpx_bit_depth_t g_bit_depth
Bit-depth of the codec.
Definition: vpx_encoder.h:375
Provides definitions for using VP8 or VP9 encoder algorithm within the vpx Codec Interface.
#define vpx_codec_enc_init(ctx, iface, cfg, flags)
Convenience macro for vpx_codec_enc_init_ver()
Definition: vpx_encoder.h:813
Codec control function to set encoder screen content mode.
Definition: vp8cx.h:324
unsigned int rc_resize_allowed
Enable/disable spatial resampling, if supported by the codec.
Definition: vpx_encoder.h:462
Bypass mode. Used when application needs to control temporal layering. This will only work when the n...
Definition: vp8cx.h:586
vpx_codec_err_t
Algorithm return codes.
Definition: vpx_codec.h:89
const vpx_codec_cx_pkt_t * vpx_codec_get_cx_data(vpx_codec_ctx_t *ctx, vpx_codec_iter_t *iter)
Encoded data iterator.
union vpx_codec_cx_pkt::@1 data
int temporal_layering_mode
Temporal layering mode indicating which temporal layering scheme to use.
Definition: vpx_encoder.h:763
int temporal_layer_id
Definition: vp8cx.h:680
Codec control function to enable/disable periodic Q boost.
Definition: vp8cx.h:424
vpx_codec_err_t vpx_codec_enc_config_default(vpx_codec_iface_t *iface, vpx_codec_enc_cfg_t *cfg, unsigned int reserved)
Get a default configuration.
#define VPX_TS_MAX_PERIODICITY
Definition: vpx_encoder.h:37
Codec control function to turn on/off SVC in encoder.
Definition: vp8cx.h:441
#define vpx_codec_control(ctx, id, data)
vpx_codec_control wrapper macro
Definition: vpx_codec.h:407
unsigned int ts_periodicity
Length of the sequence defining frame temporal layer membership.
Definition: vpx_encoder.h:738
#define VP8_EFLAG_NO_REF_ARF
Don't reference the alternate reference frame.
Definition: vp8cx.h:85
vpx_codec_err_t vpx_codec_destroy(vpx_codec_ctx_t *ctx)
Destroy a codec instance.
Definition: vpx_codec.h:220
int scaling_factor_den[12]
Definition: vpx_encoder.h:775
Codec control function to set the threshold for MBs treated static.
Definition: vp8cx.h:200
#define VPX_FRAME_IS_KEY
Definition: vpx_encoder.h:130
Definition: vpx_codec.h:221
#define VPX_EFLAG_FORCE_KF
Definition: vpx_encoder.h:305
const void * vpx_codec_iter_t
Iterator.
Definition: vpx_codec.h:188
Definition: vpx_encoder.h:176
int max_quantizers[12]
Definition: vpx_encoder.h:772
vp9 svc extra configure parameters
Definition: vpx_encoder.h:771
vpx_codec_er_flags_t g_error_resilient
Enable error resilient modes.
Definition: vpx_encoder.h:406
#define VP8_EFLAG_NO_UPD_ARF
Don't update the alternate reference frame.
Definition: vp8cx.h:109
#define VP8_EFLAG_NO_UPD_ENTROPY
Disable entropy update.
Definition: vp8cx.h:133
Codec control function to set svc layer for spatial and temporal.
Definition: vp8cx.h:458
enum vpx_rc_mode rc_end_usage
Rate control algorithm to use.
Definition: vpx_encoder.h:504
Codec context structure.
Definition: vpx_codec.h:199