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 "./y4minput.h"
23 #include "../vpx_ports/vpx_timer.h"
24 #include "vpx/vp8cx.h"
25 #include "vpx/vpx_encoder.h"
26 #include "vpx_ports/bitops.h"
27 
28 #include "../tools_common.h"
29 #include "../video_writer.h"
30 
31 #define ROI_MAP 0
32 
33 #define zero(Dest) memset(&(Dest), 0, sizeof(Dest))
34 
35 static const char *exec_name;
36 
37 void usage_exit(void) { exit(EXIT_FAILURE); }
38 
39 // Denoiser states for vp8, for temporal denoising.
40 enum denoiserStateVp8 {
41  kVp8DenoiserOff,
42  kVp8DenoiserOnYOnly,
43  kVp8DenoiserOnYUV,
44  kVp8DenoiserOnYUVAggressive,
45  kVp8DenoiserOnAdaptive
46 };
47 
48 // Denoiser states for vp9, for temporal denoising.
49 enum denoiserStateVp9 {
50  kVp9DenoiserOff,
51  kVp9DenoiserOnYOnly,
52  // For SVC: denoise the top two spatial layers.
53  kVp9DenoiserOnYTwoSpatialLayers
54 };
55 
56 static int mode_to_num_layers[13] = { 1, 2, 2, 3, 3, 3, 3, 5, 2, 3, 3, 3, 3 };
57 
58 // For rate control encoding stats.
59 struct RateControlMetrics {
60  // Number of input frames per layer.
61  int layer_input_frames[VPX_TS_MAX_LAYERS];
62  // Total (cumulative) number of encoded frames per layer.
63  int layer_tot_enc_frames[VPX_TS_MAX_LAYERS];
64  // Number of encoded non-key frames per layer.
65  int layer_enc_frames[VPX_TS_MAX_LAYERS];
66  // Framerate per layer layer (cumulative).
67  double layer_framerate[VPX_TS_MAX_LAYERS];
68  // Target average frame size per layer (per-frame-bandwidth per layer).
69  double layer_pfb[VPX_TS_MAX_LAYERS];
70  // Actual average frame size per layer.
71  double layer_avg_frame_size[VPX_TS_MAX_LAYERS];
72  // Average rate mismatch per layer (|target - actual| / target).
73  double layer_avg_rate_mismatch[VPX_TS_MAX_LAYERS];
74  // Actual encoding bitrate per layer (cumulative).
75  double layer_encoding_bitrate[VPX_TS_MAX_LAYERS];
76  // Average of the short-time encoder actual bitrate.
77  // TODO(marpan): Should we add these short-time stats for each layer?
78  double avg_st_encoding_bitrate;
79  // Variance of the short-time encoder actual bitrate.
80  double variance_st_encoding_bitrate;
81  // Window (number of frames) for computing short-timee encoding bitrate.
82  int window_size;
83  // Number of window measurements.
84  int window_count;
85  int layer_target_bitrate[VPX_MAX_LAYERS];
86 };
87 
88 // Note: these rate control metrics assume only 1 key frame in the
89 // sequence (i.e., first frame only). So for temporal pattern# 7
90 // (which has key frame for every frame on base layer), the metrics
91 // computation will be off/wrong.
92 // TODO(marpan): Update these metrics to account for multiple key frames
93 // in the stream.
94 static void set_rate_control_metrics(struct RateControlMetrics *rc,
95  vpx_codec_enc_cfg_t *cfg) {
96  int i = 0;
97  // Set the layer (cumulative) framerate and the target layer (non-cumulative)
98  // per-frame-bandwidth, for the rate control encoding stats below.
99  const double framerate = cfg->g_timebase.den / cfg->g_timebase.num;
100  const int ts_number_layers = cfg->ts_number_layers;
101  rc->layer_framerate[0] = framerate / cfg->ts_rate_decimator[0];
102  rc->layer_pfb[0] =
103  1000.0 * rc->layer_target_bitrate[0] / rc->layer_framerate[0];
104  for (i = 0; i < ts_number_layers; ++i) {
105  if (i > 0) {
106  rc->layer_framerate[i] = framerate / cfg->ts_rate_decimator[i];
107  rc->layer_pfb[i] =
108  1000.0 *
109  (rc->layer_target_bitrate[i] - rc->layer_target_bitrate[i - 1]) /
110  (rc->layer_framerate[i] - rc->layer_framerate[i - 1]);
111  }
112  rc->layer_input_frames[i] = 0;
113  rc->layer_enc_frames[i] = 0;
114  rc->layer_tot_enc_frames[i] = 0;
115  rc->layer_encoding_bitrate[i] = 0.0;
116  rc->layer_avg_frame_size[i] = 0.0;
117  rc->layer_avg_rate_mismatch[i] = 0.0;
118  }
119  rc->window_count = 0;
120  rc->window_size = 15;
121  rc->avg_st_encoding_bitrate = 0.0;
122  rc->variance_st_encoding_bitrate = 0.0;
123  // Target bandwidth for the whole stream.
124  // Set to layer_target_bitrate for highest layer (total bitrate).
125  cfg->rc_target_bitrate = rc->layer_target_bitrate[ts_number_layers - 1];
126 }
127 
128 static void printout_rate_control_summary(struct RateControlMetrics *rc,
129  vpx_codec_enc_cfg_t *cfg,
130  int frame_cnt) {
131  unsigned int i = 0;
132  int tot_num_frames = 0;
133  double perc_fluctuation = 0.0;
134  printf("Total number of processed frames: %d\n\n", frame_cnt - 1);
135  printf("Rate control layer stats for %d layer(s):\n\n",
136  cfg->ts_number_layers);
137  for (i = 0; i < cfg->ts_number_layers; ++i) {
138  const int num_dropped =
139  (i > 0) ? (rc->layer_input_frames[i] - rc->layer_enc_frames[i])
140  : (rc->layer_input_frames[i] - rc->layer_enc_frames[i] - 1);
141  tot_num_frames += rc->layer_input_frames[i];
142  rc->layer_encoding_bitrate[i] = 0.001 * rc->layer_framerate[i] *
143  rc->layer_encoding_bitrate[i] /
144  tot_num_frames;
145  rc->layer_avg_frame_size[i] =
146  rc->layer_avg_frame_size[i] / rc->layer_enc_frames[i];
147  rc->layer_avg_rate_mismatch[i] =
148  100.0 * rc->layer_avg_rate_mismatch[i] / rc->layer_enc_frames[i];
149  printf("For layer#: %d \n", i);
150  printf("Bitrate (target vs actual): %d %f \n", rc->layer_target_bitrate[i],
151  rc->layer_encoding_bitrate[i]);
152  printf("Average frame size (target vs actual): %f %f \n", rc->layer_pfb[i],
153  rc->layer_avg_frame_size[i]);
154  printf("Average rate_mismatch: %f \n", rc->layer_avg_rate_mismatch[i]);
155  printf(
156  "Number of input frames, encoded (non-key) frames, "
157  "and perc dropped frames: %d %d %f \n",
158  rc->layer_input_frames[i], rc->layer_enc_frames[i],
159  100.0 * num_dropped / rc->layer_input_frames[i]);
160  printf("\n");
161  }
162  rc->avg_st_encoding_bitrate = rc->avg_st_encoding_bitrate / rc->window_count;
163  rc->variance_st_encoding_bitrate =
164  rc->variance_st_encoding_bitrate / rc->window_count -
165  (rc->avg_st_encoding_bitrate * rc->avg_st_encoding_bitrate);
166  perc_fluctuation = 100.0 * sqrt(rc->variance_st_encoding_bitrate) /
167  rc->avg_st_encoding_bitrate;
168  printf("Short-time stats, for window of %d frames: \n", rc->window_size);
169  printf("Average, rms-variance, and percent-fluct: %f %f %f \n",
170  rc->avg_st_encoding_bitrate, sqrt(rc->variance_st_encoding_bitrate),
171  perc_fluctuation);
172  if ((frame_cnt - 1) != tot_num_frames)
173  die("Error: Number of input frames not equal to output! \n");
174 }
175 
176 #if ROI_MAP
177 static void set_roi_map(const char *enc_name, vpx_codec_enc_cfg_t *cfg,
178  vpx_roi_map_t *roi) {
179  unsigned int i, j;
180  int block_size = 0;
181  uint8_t is_vp8 = strncmp(enc_name, "vp8", 3) == 0 ? 1 : 0;
182  uint8_t is_vp9 = strncmp(enc_name, "vp9", 3) == 0 ? 1 : 0;
183  if (!is_vp8 && !is_vp9) {
184  die("unsupported codec.");
185  }
186  zero(*roi);
187 
188  block_size = is_vp9 && !is_vp8 ? 8 : 16;
189 
190  // ROI is based on the segments (4 for vp8, 8 for vp9), smallest unit for
191  // segment is 16x16 for vp8, 8x8 for vp9.
192  roi->rows = (cfg->g_h + block_size - 1) / block_size;
193  roi->cols = (cfg->g_w + block_size - 1) / block_size;
194 
195  // Applies delta QP on the segment blocks, varies from -63 to 63.
196  // Setting to negative means lower QP (better quality).
197  // Below we set delta_q to the extreme (-63) to show strong effect.
198  // VP8 uses the first 4 segments. VP9 uses all 8 segments.
199  zero(roi->delta_q);
200  roi->delta_q[1] = -63;
201 
202  // Applies delta loopfilter strength on the segment blocks, varies from -63 to
203  // 63. Setting to positive means stronger loopfilter. VP8 uses the first 4
204  // segments. VP9 uses all 8 segments.
205  zero(roi->delta_lf);
206 
207  if (is_vp8) {
208  // Applies skip encoding threshold on the segment blocks, varies from 0 to
209  // UINT_MAX. Larger value means more skipping of encoding is possible.
210  // This skip threshold only applies on delta frames.
211  zero(roi->static_threshold);
212  }
213 
214  if (is_vp9) {
215  // Apply skip segment. Setting to 1 means this block will be copied from
216  // previous frame.
217  zero(roi->skip);
218  }
219 
220  if (is_vp9) {
221  // Apply ref frame segment.
222  // -1 : Do not apply this segment.
223  // 0 : Froce using intra.
224  // 1 : Force using last.
225  // 2 : Force using golden.
226  // 3 : Force using alfref but not used in non-rd pickmode for 0 lag.
227  memset(roi->ref_frame, -1, sizeof(roi->ref_frame));
228  roi->ref_frame[1] = 1;
229  }
230 
231  // Use 2 states: 1 is center square, 0 is the rest.
232  roi->roi_map =
233  (uint8_t *)calloc(roi->rows * roi->cols, sizeof(*roi->roi_map));
234  for (i = 0; i < roi->rows; ++i) {
235  for (j = 0; j < roi->cols; ++j) {
236  if (i > (roi->rows >> 2) && i < ((roi->rows * 3) >> 2) &&
237  j > (roi->cols >> 2) && j < ((roi->cols * 3) >> 2)) {
238  roi->roi_map[i * roi->cols + j] = 1;
239  }
240  }
241  }
242 }
243 
244 static void set_roi_skip_map(vpx_codec_enc_cfg_t *cfg, vpx_roi_map_t *roi,
245  int *skip_map, int *prev_mask_map, int frame_num) {
246  const int block_size = 8;
247  unsigned int i, j;
248  roi->rows = (cfg->g_h + block_size - 1) / block_size;
249  roi->cols = (cfg->g_w + block_size - 1) / block_size;
250  zero(roi->skip);
251  zero(roi->delta_q);
252  zero(roi->delta_lf);
253  memset(roi->ref_frame, -1, sizeof(roi->ref_frame));
254  roi->ref_frame[1] = 1;
255  // Use segment 3 for skip.
256  roi->skip[3] = 1;
257  roi->roi_map =
258  (uint8_t *)calloc(roi->rows * roi->cols, sizeof(*roi->roi_map));
259  for (i = 0; i < roi->rows; ++i) {
260  for (j = 0; j < roi->cols; ++j) {
261  const int idx = i * roi->cols + j;
262  // Use segment 3 for skip.
263  // prev_mask_map keeps track of blocks that have been stably on segment 3
264  // for the past 10 frames. Only skip when the block is on segment 3 in
265  // both current map and prev_mask_map.
266  if (skip_map[idx] == 1 && prev_mask_map[idx] == 1) roi->roi_map[idx] = 3;
267  // Reset it every 10 frames so it doesn't propagate for too many frames.
268  if (frame_num % 10 == 0)
269  prev_mask_map[idx] = skip_map[idx];
270  else if (prev_mask_map[idx] == 1 && skip_map[idx] == 0)
271  prev_mask_map[idx] = 0;
272  }
273  }
274 }
275 #endif
276 
277 // Temporal scaling parameters:
278 // NOTE: The 3 prediction frames cannot be used interchangeably due to
279 // differences in the way they are handled throughout the code. The
280 // frames should be allocated to layers in the order LAST, GF, ARF.
281 // Other combinations work, but may produce slightly inferior results.
282 static void set_temporal_layer_pattern(int layering_mode,
283  vpx_codec_enc_cfg_t *cfg,
284  int *layer_flags,
285  int *flag_periodicity) {
286  switch (layering_mode) {
287  case 0: {
288  // 1-layer.
289  int ids[1] = { 0 };
290  cfg->ts_periodicity = 1;
291  *flag_periodicity = 1;
292  cfg->ts_number_layers = 1;
293  cfg->ts_rate_decimator[0] = 1;
294  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
295  // Update L only.
296  layer_flags[0] =
298  break;
299  }
300  case 1: {
301  // 2-layers, 2-frame period.
302  int ids[2] = { 0, 1 };
303  cfg->ts_periodicity = 2;
304  *flag_periodicity = 2;
305  cfg->ts_number_layers = 2;
306  cfg->ts_rate_decimator[0] = 2;
307  cfg->ts_rate_decimator[1] = 1;
308  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
309 #if 1
310  // 0=L, 1=GF, Intra-layer prediction enabled.
311  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_UPD_GF |
314  layer_flags[1] =
316 #else
317  // 0=L, 1=GF, Intra-layer prediction disabled.
318  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_UPD_GF |
321  layer_flags[1] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST |
323 #endif
324  break;
325  }
326  case 2: {
327  // 2-layers, 3-frame period.
328  int ids[3] = { 0, 1, 1 };
329  cfg->ts_periodicity = 3;
330  *flag_periodicity = 3;
331  cfg->ts_number_layers = 2;
332  cfg->ts_rate_decimator[0] = 3;
333  cfg->ts_rate_decimator[1] = 1;
334  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
335  // 0=L, 1=GF, Intra-layer prediction enabled.
336  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_REF_GF |
339  layer_flags[1] = layer_flags[2] =
342  break;
343  }
344  case 3: {
345  // 3-layers, 6-frame period.
346  int ids[6] = { 0, 2, 2, 1, 2, 2 };
347  cfg->ts_periodicity = 6;
348  *flag_periodicity = 6;
349  cfg->ts_number_layers = 3;
350  cfg->ts_rate_decimator[0] = 6;
351  cfg->ts_rate_decimator[1] = 3;
352  cfg->ts_rate_decimator[2] = 1;
353  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
354  // 0=L, 1=GF, 2=ARF, Intra-layer prediction enabled.
355  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_REF_GF |
358  layer_flags[3] =
360  layer_flags[1] = layer_flags[2] = layer_flags[4] = layer_flags[5] =
362  break;
363  }
364  case 4: {
365  // 3-layers, 4-frame period.
366  int ids[4] = { 0, 2, 1, 2 };
367  cfg->ts_periodicity = 4;
368  *flag_periodicity = 4;
369  cfg->ts_number_layers = 3;
370  cfg->ts_rate_decimator[0] = 4;
371  cfg->ts_rate_decimator[1] = 2;
372  cfg->ts_rate_decimator[2] = 1;
373  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
374  // 0=L, 1=GF, 2=ARF, Intra-layer prediction disabled.
375  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_REF_GF |
378  layer_flags[2] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
380  layer_flags[1] = layer_flags[3] =
383  break;
384  }
385  case 5: {
386  // 3-layers, 4-frame period.
387  int ids[4] = { 0, 2, 1, 2 };
388  cfg->ts_periodicity = 4;
389  *flag_periodicity = 4;
390  cfg->ts_number_layers = 3;
391  cfg->ts_rate_decimator[0] = 4;
392  cfg->ts_rate_decimator[1] = 2;
393  cfg->ts_rate_decimator[2] = 1;
394  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
395  // 0=L, 1=GF, 2=ARF, Intra-layer prediction enabled in layer 1, disabled
396  // in layer 2.
397  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_REF_GF |
400  layer_flags[2] =
402  layer_flags[1] = layer_flags[3] =
405  break;
406  }
407  case 6: {
408  // 3-layers, 4-frame period.
409  int ids[4] = { 0, 2, 1, 2 };
410  cfg->ts_periodicity = 4;
411  *flag_periodicity = 4;
412  cfg->ts_number_layers = 3;
413  cfg->ts_rate_decimator[0] = 4;
414  cfg->ts_rate_decimator[1] = 2;
415  cfg->ts_rate_decimator[2] = 1;
416  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
417  // 0=L, 1=GF, 2=ARF, Intra-layer prediction enabled.
418  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_REF_GF |
421  layer_flags[2] =
423  layer_flags[1] = layer_flags[3] =
425  break;
426  }
427  case 7: {
428  // NOTE: Probably of academic interest only.
429  // 5-layers, 16-frame period.
430  int ids[16] = { 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4 };
431  cfg->ts_periodicity = 16;
432  *flag_periodicity = 16;
433  cfg->ts_number_layers = 5;
434  cfg->ts_rate_decimator[0] = 16;
435  cfg->ts_rate_decimator[1] = 8;
436  cfg->ts_rate_decimator[2] = 4;
437  cfg->ts_rate_decimator[3] = 2;
438  cfg->ts_rate_decimator[4] = 1;
439  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
440  layer_flags[0] = VPX_EFLAG_FORCE_KF;
441  layer_flags[1] = layer_flags[3] = layer_flags[5] = layer_flags[7] =
442  layer_flags[9] = layer_flags[11] = layer_flags[13] = layer_flags[15] =
445  layer_flags[2] = layer_flags[6] = layer_flags[10] = layer_flags[14] =
447  layer_flags[4] = layer_flags[12] =
449  layer_flags[8] = VP8_EFLAG_NO_REF_LAST | VP8_EFLAG_NO_REF_GF;
450  break;
451  }
452  case 8: {
453  // 2-layers, with sync point at first frame of layer 1.
454  int ids[2] = { 0, 1 };
455  cfg->ts_periodicity = 2;
456  *flag_periodicity = 8;
457  cfg->ts_number_layers = 2;
458  cfg->ts_rate_decimator[0] = 2;
459  cfg->ts_rate_decimator[1] = 1;
460  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
461  // 0=L, 1=GF.
462  // ARF is used as predictor for all frames, and is only updated on
463  // key frame. Sync point every 8 frames.
464 
465  // Layer 0: predict from L and ARF, update L and G.
466  layer_flags[0] =
468  // Layer 1: sync point: predict from L and ARF, and update G.
469  layer_flags[1] =
471  // Layer 0, predict from L and ARF, update L.
472  layer_flags[2] =
474  // Layer 1: predict from L, G and ARF, and update G.
475  layer_flags[3] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST |
477  // Layer 0.
478  layer_flags[4] = layer_flags[2];
479  // Layer 1.
480  layer_flags[5] = layer_flags[3];
481  // Layer 0.
482  layer_flags[6] = layer_flags[4];
483  // Layer 1.
484  layer_flags[7] = layer_flags[5];
485  break;
486  }
487  case 9: {
488  // 3-layers: Sync points for layer 1 and 2 every 8 frames.
489  int ids[4] = { 0, 2, 1, 2 };
490  cfg->ts_periodicity = 4;
491  *flag_periodicity = 8;
492  cfg->ts_number_layers = 3;
493  cfg->ts_rate_decimator[0] = 4;
494  cfg->ts_rate_decimator[1] = 2;
495  cfg->ts_rate_decimator[2] = 1;
496  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
497  // 0=L, 1=GF, 2=ARF.
498  layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_REF_GF |
501  layer_flags[1] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
503  layer_flags[2] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
505  layer_flags[3] = layer_flags[5] =
507  layer_flags[4] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
509  layer_flags[6] =
511  layer_flags[7] = VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF |
513  break;
514  }
515  case 10: {
516  // 3-layers structure where ARF is used as predictor for all frames,
517  // and is only updated on key frame.
518  // Sync points for layer 1 and 2 every 8 frames.
519 
520  int ids[4] = { 0, 2, 1, 2 };
521  cfg->ts_periodicity = 4;
522  *flag_periodicity = 8;
523  cfg->ts_number_layers = 3;
524  cfg->ts_rate_decimator[0] = 4;
525  cfg->ts_rate_decimator[1] = 2;
526  cfg->ts_rate_decimator[2] = 1;
527  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
528  // 0=L, 1=GF, 2=ARF.
529  // Layer 0: predict from L and ARF; update L and G.
530  layer_flags[0] =
532  // Layer 2: sync point: predict from L and ARF; update none.
533  layer_flags[1] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_UPD_GF |
536  // Layer 1: sync point: predict from L and ARF; update G.
537  layer_flags[2] =
539  // Layer 2: predict from L, G, ARF; update none.
540  layer_flags[3] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
542  // Layer 0: predict from L and ARF; update L.
543  layer_flags[4] =
545  // Layer 2: predict from L, G, ARF; update none.
546  layer_flags[5] = layer_flags[3];
547  // Layer 1: predict from L, G, ARF; update G.
548  layer_flags[6] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST;
549  // Layer 2: predict from L, G, ARF; update none.
550  layer_flags[7] = layer_flags[3];
551  break;
552  }
553  case 11: {
554  // 3-layers structure with one reference frame.
555  // This works same as temporal_layering_mode 3.
556  // This was added to compare with vp9_spatial_svc_encoder.
557 
558  // 3-layers, 4-frame period.
559  int ids[4] = { 0, 2, 1, 2 };
560  cfg->ts_periodicity = 4;
561  *flag_periodicity = 4;
562  cfg->ts_number_layers = 3;
563  cfg->ts_rate_decimator[0] = 4;
564  cfg->ts_rate_decimator[1] = 2;
565  cfg->ts_rate_decimator[2] = 1;
566  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
567  // 0=L, 1=GF, 2=ARF, Intra-layer prediction disabled.
568  layer_flags[0] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
570  layer_flags[2] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
572  layer_flags[1] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
574  layer_flags[3] = VP8_EFLAG_NO_REF_LAST | VP8_EFLAG_NO_REF_ARF |
576  break;
577  }
578  case 12:
579  default: {
580  // 3-layers structure as in case 10, but no sync/refresh points for
581  // layer 1 and 2.
582  int ids[4] = { 0, 2, 1, 2 };
583  cfg->ts_periodicity = 4;
584  *flag_periodicity = 8;
585  cfg->ts_number_layers = 3;
586  cfg->ts_rate_decimator[0] = 4;
587  cfg->ts_rate_decimator[1] = 2;
588  cfg->ts_rate_decimator[2] = 1;
589  memcpy(cfg->ts_layer_id, ids, sizeof(ids));
590  // 0=L, 1=GF, 2=ARF.
591  // Layer 0: predict from L and ARF; update L.
592  layer_flags[0] =
594  layer_flags[4] = layer_flags[0];
595  // Layer 1: predict from L, G, ARF; update G.
596  layer_flags[2] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST;
597  layer_flags[6] = layer_flags[2];
598  // Layer 2: predict from L, G, ARF; update none.
599  layer_flags[1] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
601  layer_flags[3] = layer_flags[1];
602  layer_flags[5] = layer_flags[1];
603  layer_flags[7] = layer_flags[1];
604  break;
605  }
606  }
607 }
608 
609 #if ROI_MAP
610 static void read_mask(FILE *mask_file, int *seg_map) {
611  int mask_rows, mask_cols, i, j;
612  int *map_start = seg_map;
613  fscanf(mask_file, "%d %d\n", &mask_cols, &mask_rows);
614  for (i = 0; i < mask_rows; i++) {
615  for (j = 0; j < mask_cols; j++) {
616  fscanf(mask_file, "%d ", &seg_map[j]);
617  // reverse the bit
618  seg_map[j] = 1 - seg_map[j];
619  }
620  seg_map += mask_cols;
621  }
622  seg_map = map_start;
623 }
624 #endif
625 
626 int main(int argc, char **argv) {
627  VpxVideoWriter *outfile[VPX_TS_MAX_LAYERS] = { NULL };
628  vpx_codec_ctx_t codec;
630  int frame_cnt = 0;
631  vpx_image_t raw;
632  vpx_codec_err_t res;
633  unsigned int width;
634  unsigned int height;
635  uint32_t error_resilient = 0;
636  int speed;
637  int frame_avail;
638  int got_data;
639  int flags = 0;
640  unsigned int i;
641  int pts = 0; // PTS starts at 0.
642  int frame_duration = 1; // 1 timebase tick per frame.
643  int layering_mode = 0;
644  int layer_flags[VPX_TS_MAX_PERIODICITY] = { 0 };
645  int flag_periodicity = 1;
646 #if ROI_MAP
647  vpx_roi_map_t roi;
648 #endif
649  vpx_svc_layer_id_t layer_id;
650  const VpxInterface *encoder = NULL;
651  struct VpxInputContext input_ctx;
652  struct RateControlMetrics rc;
653  int64_t cx_time = 0;
654  const int min_args_base = 13;
655 #if CONFIG_VP9_HIGHBITDEPTH
656  vpx_bit_depth_t bit_depth = VPX_BITS_8;
657  int input_bit_depth = 8;
658  const int min_args = min_args_base + 1;
659 #else
660  const int min_args = min_args_base;
661 #endif // CONFIG_VP9_HIGHBITDEPTH
662  double sum_bitrate = 0.0;
663  double sum_bitrate2 = 0.0;
664  double framerate = 30.0;
665 #if ROI_MAP
666  FILE *mask_file = NULL;
667  int block_size = 8;
668  int mask_rows = 0;
669  int mask_cols = 0;
670  int *mask_map;
671  int *prev_mask_map;
672 #endif
673  zero(rc.layer_target_bitrate);
674  memset(&layer_id, 0, sizeof(vpx_svc_layer_id_t));
675  memset(&input_ctx, 0, sizeof(input_ctx));
676  /* Setup default input stream settings */
677  input_ctx.framerate.numerator = 30;
678  input_ctx.framerate.denominator = 1;
679  input_ctx.only_i420 = 1;
680  input_ctx.bit_depth = 0;
681 
682  exec_name = argv[0];
683  // Check usage and arguments.
684  if (argc < min_args) {
685 #if CONFIG_VP9_HIGHBITDEPTH
686  die("Usage: %s <infile> <outfile> <codec_type(vp8/vp9)> <width> <height> "
687  "<rate_num> <rate_den> <speed> <frame_drop_threshold> "
688  "<error_resilient> <threads> <mode> "
689  "<Rate_0> ... <Rate_nlayers-1> <bit-depth> \n",
690  argv[0]);
691 #else
692  die("Usage: %s <infile> <outfile> <codec_type(vp8/vp9)> <width> <height> "
693  "<rate_num> <rate_den> <speed> <frame_drop_threshold> "
694  "<error_resilient> <threads> <mode> "
695  "<Rate_0> ... <Rate_nlayers-1> \n",
696  argv[0]);
697 #endif // CONFIG_VP9_HIGHBITDEPTH
698  }
699 
700  encoder = get_vpx_encoder_by_name(argv[3]);
701  if (!encoder) die("Unsupported codec.");
702 
703  printf("Using %s\n", vpx_codec_iface_name(encoder->codec_interface()));
704 
705  width = (unsigned int)strtoul(argv[4], NULL, 0);
706  height = (unsigned int)strtoul(argv[5], NULL, 0);
707  if (width < 16 || width % 2 || height < 16 || height % 2) {
708  die("Invalid resolution: %d x %d", width, height);
709  }
710 
711  layering_mode = (int)strtol(argv[12], NULL, 0);
712  if (layering_mode < 0 || layering_mode > 13) {
713  die("Invalid layering mode (0..12) %s", argv[12]);
714  }
715 
716 #if ROI_MAP
717  if (argc != min_args + mode_to_num_layers[layering_mode] + 1) {
718  die("Invalid number of arguments");
719  }
720 #else
721  if (argc != min_args + mode_to_num_layers[layering_mode]) {
722  die("Invalid number of arguments");
723  }
724 #endif
725 
726  input_ctx.filename = argv[1];
727  open_input_file(&input_ctx);
728 
729 #if CONFIG_VP9_HIGHBITDEPTH
730  switch (strtol(argv[argc - 1], NULL, 0)) {
731  case 8:
732  bit_depth = VPX_BITS_8;
733  input_bit_depth = 8;
734  break;
735  case 10:
736  bit_depth = VPX_BITS_10;
737  input_bit_depth = 10;
738  break;
739  case 12:
740  bit_depth = VPX_BITS_12;
741  input_bit_depth = 12;
742  break;
743  default: die("Invalid bit depth (8, 10, 12) %s", argv[argc - 1]);
744  }
745 
746  // Y4M reader has its own allocation.
747  if (input_ctx.file_type != FILE_TYPE_Y4M) {
748  if (!vpx_img_alloc(
749  &raw,
751  width, height, 32)) {
752  die("Failed to allocate image (%dx%d)", width, height);
753  }
754  }
755 #else
756  // Y4M reader has its own allocation.
757  if (input_ctx.file_type != FILE_TYPE_Y4M) {
758  if (!vpx_img_alloc(&raw, VPX_IMG_FMT_I420, width, height, 32)) {
759  die("Failed to allocate image (%dx%d)", width, height);
760  }
761  }
762 #endif // CONFIG_VP9_HIGHBITDEPTH
763 
764  // Populate encoder configuration.
765  res = vpx_codec_enc_config_default(encoder->codec_interface(), &cfg, 0);
766  if (res) {
767  printf("Failed to get config: %s\n", vpx_codec_err_to_string(res));
768  return EXIT_FAILURE;
769  }
770 
771  // Update the default configuration with our settings.
772  cfg.g_w = width;
773  cfg.g_h = height;
774 
775 #if CONFIG_VP9_HIGHBITDEPTH
776  if (bit_depth != VPX_BITS_8) {
777  cfg.g_bit_depth = bit_depth;
778  cfg.g_input_bit_depth = input_bit_depth;
779  cfg.g_profile = 2;
780  }
781 #endif // CONFIG_VP9_HIGHBITDEPTH
782 
783  // Timebase format e.g. 30fps: numerator=1, demoninator = 30.
784  cfg.g_timebase.num = (int)strtol(argv[6], NULL, 0);
785  cfg.g_timebase.den = (int)strtol(argv[7], NULL, 0);
786 
787  speed = (int)strtol(argv[8], NULL, 0);
788  if (speed < 0) {
789  die("Invalid speed setting: must be positive");
790  }
791  if (strncmp(encoder->name, "vp9", 3) == 0 && speed > 9) {
792  warn("Mapping speed %d to speed 9.\n", speed);
793  }
794 
795  for (i = min_args_base;
796  (int)i < min_args_base + mode_to_num_layers[layering_mode]; ++i) {
797  rc.layer_target_bitrate[i - 13] = (int)strtol(argv[i], NULL, 0);
798  if (strncmp(encoder->name, "vp8", 3) == 0)
799  cfg.ts_target_bitrate[i - 13] = rc.layer_target_bitrate[i - 13];
800  else if (strncmp(encoder->name, "vp9", 3) == 0)
801  cfg.layer_target_bitrate[i - 13] = rc.layer_target_bitrate[i - 13];
802  }
803 
804  // Real time parameters.
805  cfg.rc_dropframe_thresh = (unsigned int)strtoul(argv[9], NULL, 0);
806  cfg.rc_end_usage = VPX_CBR;
807  cfg.rc_min_quantizer = 2;
808  cfg.rc_max_quantizer = 56;
809  if (strncmp(encoder->name, "vp9", 3) == 0) cfg.rc_max_quantizer = 52;
810  cfg.rc_undershoot_pct = 50;
811  cfg.rc_overshoot_pct = 50;
812  cfg.rc_buf_initial_sz = 600;
813  cfg.rc_buf_optimal_sz = 600;
814  cfg.rc_buf_sz = 1000;
815 
816  // Disable dynamic resizing by default.
817  cfg.rc_resize_allowed = 0;
818 
819  // Use 1 thread as default.
820  cfg.g_threads = (unsigned int)strtoul(argv[11], NULL, 0);
821 
822  error_resilient = (uint32_t)strtoul(argv[10], NULL, 0);
823  if (error_resilient != 0 && error_resilient != 1) {
824  die("Invalid value for error resilient (0, 1): %d.", error_resilient);
825  }
826  // Enable error resilient mode.
827  cfg.g_error_resilient = error_resilient;
828  cfg.g_lag_in_frames = 0;
829  cfg.kf_mode = VPX_KF_AUTO;
830 
831  // Disable automatic keyframe placement.
832  cfg.kf_min_dist = cfg.kf_max_dist = 3000;
833 
835 
836  set_temporal_layer_pattern(layering_mode, &cfg, layer_flags,
837  &flag_periodicity);
838 
839  set_rate_control_metrics(&rc, &cfg);
840 
841  if (input_ctx.file_type == FILE_TYPE_Y4M) {
842  if (input_ctx.width != cfg.g_w || input_ctx.height != cfg.g_h) {
843  die("Incorrect width or height: %d x %d", cfg.g_w, cfg.g_h);
844  }
845  if (input_ctx.framerate.numerator != cfg.g_timebase.den ||
846  input_ctx.framerate.denominator != cfg.g_timebase.num) {
847  die("Incorrect framerate: numerator %d denominator %d",
848  cfg.g_timebase.num, cfg.g_timebase.den);
849  }
850  }
851 
852  framerate = cfg.g_timebase.den / cfg.g_timebase.num;
853  // Open an output file for each stream.
854  for (i = 0; i < cfg.ts_number_layers; ++i) {
855  char file_name[PATH_MAX];
856  VpxVideoInfo info;
857  info.codec_fourcc = encoder->fourcc;
858  info.frame_width = cfg.g_w;
859  info.frame_height = cfg.g_h;
860  info.time_base.numerator = cfg.g_timebase.num;
861  info.time_base.denominator = cfg.g_timebase.den;
862 
863  snprintf(file_name, sizeof(file_name), "%s_%d.ivf", argv[2], i);
864  outfile[i] = vpx_video_writer_open(file_name, kContainerIVF, &info);
865  if (!outfile[i]) die("Failed to open %s for writing", file_name);
866 
867  assert(outfile[i] != NULL);
868  }
869  // No spatial layers in this encoder.
870  cfg.ss_number_layers = 1;
871 
872 // Initialize codec.
873 #if CONFIG_VP9_HIGHBITDEPTH
874  if (vpx_codec_enc_init(
875  &codec, encoder->codec_interface(), &cfg,
876  bit_depth == VPX_BITS_8 ? 0 : VPX_CODEC_USE_HIGHBITDEPTH))
877 #else
878  if (vpx_codec_enc_init(&codec, encoder->codec_interface(), &cfg, 0))
879 #endif // CONFIG_VP9_HIGHBITDEPTH
880  die("Failed to initialize encoder");
881 
882 #if ROI_MAP
883  mask_rows = (cfg.g_h + block_size - 1) / block_size;
884  mask_cols = (cfg.g_w + block_size - 1) / block_size;
885  mask_map = (int *)calloc(mask_rows * mask_cols, sizeof(*mask_map));
886  prev_mask_map = (int *)calloc(mask_rows * mask_cols, sizeof(*mask_map));
887 #endif
888 
889  if (strncmp(encoder->name, "vp8", 3) == 0) {
890  vpx_codec_control(&codec, VP8E_SET_CPUUSED, -speed);
891  vpx_codec_control(&codec, VP8E_SET_NOISE_SENSITIVITY, kVp8DenoiserOff);
894 #if ROI_MAP
895  set_roi_map(encoder->name, &cfg, &roi);
896  if (vpx_codec_control(&codec, VP8E_SET_ROI_MAP, &roi))
897  die_codec(&codec, "Failed to set ROI map");
898 #endif
899  } else if (strncmp(encoder->name, "vp9", 3) == 0) {
900  vpx_svc_extra_cfg_t svc_params;
901  memset(&svc_params, 0, sizeof(svc_params));
904  vpx_codec_control(&codec, VP8E_SET_CPUUSED, speed);
909  vpx_codec_control(&codec, VP9E_SET_NOISE_SENSITIVITY, kVp9DenoiserOff);
912  vpx_codec_control(&codec, VP9E_SET_TILE_COLUMNS, get_msb(cfg.g_threads));
914 
915  if (cfg.g_threads > 1)
917  else
919  if (vpx_codec_control(&codec, VP9E_SET_SVC, layering_mode > 0 ? 1 : 0))
920  die_codec(&codec, "Failed to set SVC");
921  for (i = 0; i < cfg.ts_number_layers; ++i) {
922  svc_params.max_quantizers[i] = cfg.rc_max_quantizer;
923  svc_params.min_quantizers[i] = cfg.rc_min_quantizer;
924  }
925  svc_params.scaling_factor_num[0] = cfg.g_h;
926  svc_params.scaling_factor_den[0] = cfg.g_h;
927  vpx_codec_control(&codec, VP9E_SET_SVC_PARAMETERS, &svc_params);
928  }
929  if (strncmp(encoder->name, "vp8", 3) == 0) {
931  }
933  // This controls the maximum target size of the key frame.
934  // For generating smaller key frames, use a smaller max_intra_size_pct
935  // value, like 100 or 200.
936  {
937  const int max_intra_size_pct = 1000;
939  max_intra_size_pct);
940  }
941 
942  frame_avail = 1;
943  while (frame_avail || got_data) {
944  struct vpx_usec_timer timer;
945  vpx_codec_iter_t iter = NULL;
946  const vpx_codec_cx_pkt_t *pkt;
947 #if ROI_MAP
948  char mask_file_name[255];
949 #endif
950  // Update the temporal layer_id. No spatial layers in this test.
951  layer_id.spatial_layer_id = 0;
952  layer_id.temporal_layer_id =
953  cfg.ts_layer_id[frame_cnt % cfg.ts_periodicity];
954  layer_id.temporal_layer_id_per_spatial[0] = layer_id.temporal_layer_id;
955  if (strncmp(encoder->name, "vp9", 3) == 0) {
956  vpx_codec_control(&codec, VP9E_SET_SVC_LAYER_ID, &layer_id);
957  } else if (strncmp(encoder->name, "vp8", 3) == 0) {
959  layer_id.temporal_layer_id);
960  }
961  flags = layer_flags[frame_cnt % flag_periodicity];
962  if (layering_mode == 0) flags = 0;
963 #if ROI_MAP
964  snprintf(mask_file_name, sizeof(mask_file_name), "%s%05d.txt",
965  argv[argc - 1], frame_cnt);
966  mask_file = fopen(mask_file_name, "r");
967  if (mask_file != NULL) {
968  read_mask(mask_file, mask_map);
969  fclose(mask_file);
970  // set_roi_map(encoder->name, &cfg, &roi);
971  set_roi_skip_map(&cfg, &roi, mask_map, prev_mask_map, frame_cnt);
972  if (vpx_codec_control(&codec, VP9E_SET_ROI_MAP, &roi))
973  die_codec(&codec, "Failed to set ROI map");
974  }
975 #endif
976  frame_avail = read_frame(&input_ctx, &raw);
977  if (frame_avail) ++rc.layer_input_frames[layer_id.temporal_layer_id];
978  vpx_usec_timer_start(&timer);
979  if (vpx_codec_encode(&codec, frame_avail ? &raw : NULL, pts, 1, flags,
980  VPX_DL_REALTIME)) {
981  die_codec(&codec, "Failed to encode frame");
982  }
983  vpx_usec_timer_mark(&timer);
984  cx_time += vpx_usec_timer_elapsed(&timer);
985  // Reset KF flag.
986  if (layering_mode != 7) {
987  layer_flags[0] &= ~VPX_EFLAG_FORCE_KF;
988  }
989  got_data = 0;
990  while ((pkt = vpx_codec_get_cx_data(&codec, &iter))) {
991  got_data = 1;
992  switch (pkt->kind) {
994  for (i = cfg.ts_layer_id[frame_cnt % cfg.ts_periodicity];
995  i < cfg.ts_number_layers; ++i) {
996  vpx_video_writer_write_frame(outfile[i], pkt->data.frame.buf,
997  pkt->data.frame.sz, pts);
998  ++rc.layer_tot_enc_frames[i];
999  rc.layer_encoding_bitrate[i] += 8.0 * pkt->data.frame.sz;
1000  // Keep count of rate control stats per layer (for non-key frames).
1001  if (i == cfg.ts_layer_id[frame_cnt % cfg.ts_periodicity] &&
1002  !(pkt->data.frame.flags & VPX_FRAME_IS_KEY)) {
1003  rc.layer_avg_frame_size[i] += 8.0 * pkt->data.frame.sz;
1004  rc.layer_avg_rate_mismatch[i] +=
1005  fabs(8.0 * pkt->data.frame.sz - rc.layer_pfb[i]) /
1006  rc.layer_pfb[i];
1007  ++rc.layer_enc_frames[i];
1008  }
1009  }
1010  // Update for short-time encoding bitrate states, for moving window
1011  // of size rc->window, shifted by rc->window / 2.
1012  // Ignore first window segment, due to key frame.
1013  if (rc.window_size == 0) rc.window_size = 15;
1014  if (frame_cnt > rc.window_size) {
1015  sum_bitrate += 0.001 * 8.0 * pkt->data.frame.sz * framerate;
1016  if (frame_cnt % rc.window_size == 0) {
1017  rc.window_count += 1;
1018  rc.avg_st_encoding_bitrate += sum_bitrate / rc.window_size;
1019  rc.variance_st_encoding_bitrate +=
1020  (sum_bitrate / rc.window_size) *
1021  (sum_bitrate / rc.window_size);
1022  sum_bitrate = 0.0;
1023  }
1024  }
1025  // Second shifted window.
1026  if (frame_cnt > rc.window_size + rc.window_size / 2) {
1027  sum_bitrate2 += 0.001 * 8.0 * pkt->data.frame.sz * framerate;
1028  if (frame_cnt > 2 * rc.window_size &&
1029  frame_cnt % rc.window_size == 0) {
1030  rc.window_count += 1;
1031  rc.avg_st_encoding_bitrate += sum_bitrate2 / rc.window_size;
1032  rc.variance_st_encoding_bitrate +=
1033  (sum_bitrate2 / rc.window_size) *
1034  (sum_bitrate2 / rc.window_size);
1035  sum_bitrate2 = 0.0;
1036  }
1037  }
1038  break;
1039  default: break;
1040  }
1041  }
1042  ++frame_cnt;
1043  pts += frame_duration;
1044  }
1045 #if ROI_MAP
1046  free(mask_map);
1047  free(prev_mask_map);
1048 #endif
1049  close_input_file(&input_ctx);
1050  printout_rate_control_summary(&rc, &cfg, frame_cnt);
1051  printf("\n");
1052  printf("Frame cnt and encoding time/FPS stats for encoding: %d %f %f \n",
1053  frame_cnt, 1000 * (float)cx_time / (double)(frame_cnt * 1000000),
1054  1000000 * (double)frame_cnt / (double)cx_time);
1055 
1056  if (vpx_codec_destroy(&codec)) die_codec(&codec, "Failed to destroy codec");
1057 
1058  // Try to rewrite the output file headers with the actual frame count.
1059  for (i = 0; i < cfg.ts_number_layers; ++i) vpx_video_writer_close(outfile[i]);
1060 
1061  if (input_ctx.file_type != FILE_TYPE_Y4M) {
1062  vpx_img_free(&raw);
1063  }
1064 
1065 #if ROI_MAP
1066  free(roi.roi_map);
1067 #endif
1068  return EXIT_SUCCESS;
1069 }
vpx_codec_err_t vpx_codec_destroy(vpx_codec_ctx_t *ctx)
Destroy a codec instance.
const void * vpx_codec_iter_t
Iterator.
Definition: vpx_codec.h:190
enum vpx_bit_depth vpx_bit_depth_t
Bit depth for codecThis enumeration determines the bit depth of the codec.
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_codec_control(ctx, id, data)
vpx_codec_control wrapper macro
Definition: vpx_codec.h:407
vpx_codec_err_t
Algorithm return codes.
Definition: vpx_codec.h:93
@ VPX_BITS_8
Definition: vpx_codec.h:221
@ VPX_BITS_12
Definition: vpx_codec.h:223
@ VPX_BITS_10
Definition: vpx_codec.h:222
#define VPX_DL_REALTIME
deadline parameter analogous to VPx REALTIME mode.
Definition: vpx_encoder.h:978
#define VPX_TS_MAX_LAYERS
Definition: vpx_encoder.h:41
#define vpx_codec_enc_init(ctx, iface, cfg, flags)
Convenience macro for vpx_codec_enc_init_ver()
Definition: vpx_encoder.h:889
#define VPX_EFLAG_FORCE_KF
Definition: vpx_encoder.h:262
#define VPX_TS_MAX_PERIODICITY
Definition: vpx_encoder.h:38
#define VPX_CODEC_USE_HIGHBITDEPTH
Definition: vpx_encoder.h:92
#define VPX_MAX_LAYERS
Definition: vpx_encoder.h:44
#define VPX_FRAME_IS_KEY
Definition: vpx_encoder.h:118
vpx_codec_err_t vpx_codec_enc_config_default(vpx_codec_iface_t *iface, vpx_codec_enc_cfg_t *cfg, unsigned int usage)
Get a default configuration.
const vpx_codec_cx_pkt_t * vpx_codec_get_cx_data(vpx_codec_ctx_t *ctx, vpx_codec_iter_t *iter)
Encoded data iterator.
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.
@ VPX_CODEC_CX_FRAME_PKT
Definition: vpx_encoder.h:149
@ VPX_KF_AUTO
Definition: vpx_encoder.h:250
@ VPX_CBR
Definition: vpx_encoder.h:235
#define VP8_EFLAG_NO_UPD_ARF
Don't update the alternate reference frame.
Definition: vp8cx.h:112
#define VP8_EFLAG_NO_UPD_ENTROPY
Disable entropy update.
Definition: vp8cx.h:133
#define VP8_EFLAG_NO_UPD_LAST
Don't update the last frame.
Definition: vp8cx.h:98
#define VP8_EFLAG_NO_REF_ARF
Don't reference the alternate reference frame.
Definition: vp8cx.h:91
#define VP8_EFLAG_NO_UPD_GF
Don't update the golden frame.
Definition: vp8cx.h:105
#define VP8_EFLAG_NO_REF_GF
Don't reference the golden frame.
Definition: vp8cx.h:83
#define VP8_EFLAG_NO_REF_LAST
Don't reference the last frame.
Definition: vp8cx.h:75
@ VP9E_SET_FRAME_PERIODIC_BOOST
Codec control function to enable/disable periodic Q boost.
Definition: vp8cx.h:430
@ VP9E_SET_SVC_LAYER_ID
Codec control function to set svc layer for spatial and temporal.
Definition: vp8cx.h:470
@ VP8E_SET_MAX_INTRA_BITRATE_PCT
Codec control function to set Max data rate for Intra frames.
Definition: vp8cx.h:274
@ VP9E_SET_ROI_MAP
Codec control function to pass an ROI map to encoder.
Definition: vp8cx.h:453
@ VP8E_SET_ROI_MAP
Codec control function to pass an ROI map to encoder.
Definition: vp8cx.h:147
@ VP9E_SET_AQ_MODE
Codec control function to set adaptive quantization mode.
Definition: vp8cx.h:415
@ VP8E_SET_NOISE_SENSITIVITY
control function to set noise sensitivity
Definition: vp8cx.h:190
@ VP8E_SET_TOKEN_PARTITIONS
Codec control function to set the number of token partitions.
Definition: vp8cx.h:211
@ VP9E_SET_POSTENCODE_DROP
Codec control function to enable postencode frame drop.
Definition: vp8cx.h:683
@ VP9E_SET_DISABLE_OVERSHOOT_MAXQ_CBR
Codec control function to disable increase Q on overshoot in CBR.
Definition: vp8cx.h:699
@ VP9E_SET_SVC_PARAMETERS
Codec control function to set parameters for SVC.
Definition: vp8cx.h:461
@ VP9E_SET_FRAME_PARALLEL_DECODING
Codec control function to enable frame parallel decoding feature.
Definition: vp8cx.h:402
@ VP8E_SET_GF_CBR_BOOST_PCT
Boost percentage for Golden Frame in CBR mode.
Definition: vp8cx.h:606
@ VP9E_SET_TUNE_CONTENT
Codec control function to set content type.
Definition: vp8cx.h:480
@ VP9E_SET_SVC
Codec control function to turn on/off SVC in encoder.
Definition: vp8cx.h:447
@ VP9E_SET_ROW_MT
Codec control function to set row level multi-threading.
Definition: vp8cx.h:575
@ VP8E_SET_CPUUSED
Codec control function to set encoder internal speed settings.
Definition: vp8cx.h:172
@ VP8E_SET_TEMPORAL_LAYER_ID
Codec control function to set the temporal layer id.
Definition: vp8cx.h:321
@ VP9E_SET_TILE_COLUMNS
Codec control function to set number of tile columns.
Definition: vp8cx.h:368
@ VP8E_SET_STATIC_THRESHOLD
Codec control function to set the threshold for MBs treated static.
Definition: vp8cx.h:205
@ VP8E_SET_SCREEN_CONTENT_MODE
Codec control function to set encoder screen content mode.
Definition: vp8cx.h:329
@ VP9E_SET_DISABLE_LOOPFILTER
Codec control function to disable loopfilter.
Definition: vp8cx.h:708
@ VP9E_SET_NOISE_SENSITIVITY
Codec control function to set noise sensitivity.
Definition: vp8cx.h:438
@ VP9E_SET_GF_CBR_BOOST_PCT
Boost percentage for Golden Frame in CBR mode.
Definition: vp8cx.h:310
@ VP9E_TEMPORAL_LAYERING_MODE_BYPASS
Bypass mode. Used when application needs to control temporal layering. This will only work when the n...
Definition: vp8cx.h:789
Codec context structure.
Definition: vpx_codec.h:200
Encoder output packet.
Definition: vpx_encoder.h:161
enum vpx_codec_cx_pkt_kind kind
Definition: vpx_encoder.h:162
struct vpx_codec_cx_pkt::@1::@2 frame
union vpx_codec_cx_pkt::@1 data
Encoder configuration structure.
Definition: vpx_encoder.h:270
unsigned int rc_resize_allowed
Enable/disable spatial resampling, if supported by the codec.
Definition: vpx_encoder.h:402
int temporal_layering_mode
Temporal layering mode indicating which temporal layering scheme to use.
Definition: vpx_encoder.h:695
unsigned int kf_min_dist
Keyframe minimum interval.
Definition: vpx_encoder.h:607
unsigned int rc_min_quantizer
Minimum (Best Quality) Quantizer.
Definition: vpx_encoder.h:475
unsigned int ts_number_layers
Number of temporal coding layers.
Definition: vpx_encoder.h:646
unsigned int ss_number_layers
Number of spatial coding layers.
Definition: vpx_encoder.h:626
unsigned int g_profile
Bitstream profile to use.
Definition: vpx_encoder.h:297
unsigned int layer_target_bitrate[12]
Target bitrate for each spatial/temporal layer.
Definition: vpx_encoder.h:686
unsigned int g_h
Height of the frame.
Definition: vpx_encoder.h:315
enum vpx_kf_mode kf_mode
Keyframe placement mode.
Definition: vpx_encoder.h:598
unsigned int ts_layer_id[16]
Template defining the membership of frames to temporal layers.
Definition: vpx_encoder.h:678
vpx_codec_er_flags_t g_error_resilient
Enable error resilient modes.
Definition: vpx_encoder.h:353
unsigned int ts_periodicity
Length of the sequence defining frame temporal layer membership.
Definition: vpx_encoder.h:669
unsigned int rc_overshoot_pct
Rate control adaptation overshoot control.
Definition: vpx_encoder.h:518
unsigned int g_w
Width of the frame.
Definition: vpx_encoder.h:306
unsigned int rc_buf_sz
Decoder Buffer Size.
Definition: vpx_encoder.h:533
unsigned int rc_dropframe_thresh
Temporal resampling configuration, if supported by the codec.
Definition: vpx_encoder.h:393
struct vpx_rational g_timebase
Stream timebase units.
Definition: vpx_encoder.h:345
unsigned int rc_max_quantizer
Maximum (Worst Quality) Quantizer.
Definition: vpx_encoder.h:484
unsigned int g_lag_in_frames
Allow lagged encoding.
Definition: vpx_encoder.h:374
enum vpx_rc_mode rc_end_usage
Rate control algorithm to use.
Definition: vpx_encoder.h:442
unsigned int rc_buf_initial_sz
Decoder Buffer Initial Size.
Definition: vpx_encoder.h:542
vpx_bit_depth_t g_bit_depth
Bit-depth of the codec.
Definition: vpx_encoder.h:323
unsigned int rc_buf_optimal_sz
Decoder Buffer Optimal Size.
Definition: vpx_encoder.h:551
unsigned int rc_target_bitrate
Target data rate.
Definition: vpx_encoder.h:462
unsigned int ts_target_bitrate[5]
Target bitrate for each temporal layer.
Definition: vpx_encoder.h:653
unsigned int g_input_bit_depth
Bit-depth of the input frames.
Definition: vpx_encoder.h:331
unsigned int rc_undershoot_pct
Rate control adaptation undershoot control.
Definition: vpx_encoder.h:503
unsigned int ts_rate_decimator[5]
Frame rate decimation factor for each temporal layer.
Definition: vpx_encoder.h:660
unsigned int kf_max_dist
Keyframe maximum interval.
Definition: vpx_encoder.h:616
unsigned int g_threads
Maximum number of threads to use.
Definition: vpx_encoder.h:287
Image Descriptor.
Definition: vpx_image.h:72
int den
Definition: vpx_encoder.h:222
int num
Definition: vpx_encoder.h:221
vpx region of interest map
Definition: vp8cx.h:806
int skip[8]
Definition: vp8cx.h:818
unsigned int static_threshold[4]
Definition: vp8cx.h:821
unsigned int rows
Definition: vp8cx.h:812
unsigned int cols
Definition: vp8cx.h:813
int ref_frame[8]
Definition: vp8cx.h:819
int delta_q[8]
Definition: vp8cx.h:815
unsigned char * roi_map
Definition: vp8cx.h:811
int delta_lf[8]
Definition: vp8cx.h:816
vp9 svc layer parameters
Definition: vp8cx.h:883
int temporal_layer_id
Definition: vp8cx.h:886
int temporal_layer_id_per_spatial[5]
Definition: vp8cx.h:887
int spatial_layer_id
Definition: vp8cx.h:884
vp9 svc extra configure parameters
Definition: vpx_encoder.h:848
int min_quantizers[12]
Definition: vpx_encoder.h:850
int scaling_factor_num[12]
Definition: vpx_encoder.h:851
int max_quantizers[12]
Definition: vpx_encoder.h:849
int scaling_factor_den[12]
Definition: vpx_encoder.h:852
Provides definitions for using VP8 or VP9 encoder algorithm within the vpx Codec Interface.
Describes the encoder algorithm interface to applications.
@ VPX_IMG_FMT_I42016
Definition: vpx_image.h:47
@ VPX_IMG_FMT_I420
Definition: vpx_image.h:42
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.
void vpx_img_free(vpx_image_t *img)
Close an image descriptor.