vpx_temporal_svc_encoder.c 30.9 KB
Newer Older
1 2 3 4 5 6 7 8 9 10
/*
 *  Copyright (c) 2012 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

11
//  This is an example demonstrating how to implement a multi-layer VPx
12 13 14
//  encoding scheme based on temporal scalability for video applications
//  that benefit from a scalable bitstream.

15
#include <assert.h>
16
#include <math.h>
17 18 19 20
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

21
#include "./vpx_config.h"
22
#include "../vpx_ports/vpx_timer.h"
23 24 25
#include "vpx/vp8cx.h"
#include "vpx/vpx_encoder.h"

26 27
#include "../tools_common.h"
#include "../video_writer.h"
28

29 30
static const char *exec_name;

31
void usage_exit(void) {
32 33 34
  exit(EXIT_FAILURE);
}

35 36 37 38 39
// Denoiser states, for temporal denoising.
enum denoiserState {
  kDenoiserOff,
  kDenoiserOnYOnly,
  kDenoiserOnYUV,
40 41
  kDenoiserOnYUVAggressive,
  kDenoiserOnAdaptive
42 43
};

44 45
static int mode_to_num_layers[12] = {1, 2, 2, 3, 3, 3, 3, 5, 2, 3, 3, 3};

46 47 48 49 50 51 52 53 54
// For rate control encoding stats.
struct RateControlMetrics {
  // Number of input frames per layer.
  int layer_input_frames[VPX_TS_MAX_LAYERS];
  // Total (cumulative) number of encoded frames per layer.
  int layer_tot_enc_frames[VPX_TS_MAX_LAYERS];
  // Number of encoded non-key frames per layer.
  int layer_enc_frames[VPX_TS_MAX_LAYERS];
  // 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 65 66 67 68 69 70 71 72
  // Average of the short-time encoder actual bitrate.
  // TODO(marpan): Should we add these short-time stats for each layer?
  double avg_st_encoding_bitrate;
  // Variance of the short-time encoder actual bitrate.
  double variance_st_encoding_bitrate;
  // Window (number of frames) for computing short-timee encoding bitrate.
  int window_size;
  // Number of window measurements.
  int window_count;
73
  int layer_target_bitrate[VPX_MAX_LAYERS];
74 75
};

76 77 78 79 80 81
// Note: these rate control metrics assume only 1 key frame in the
// sequence (i.e., first frame only). So for temporal pattern# 7
// (which has key frame for every frame on base layer), the metrics
// computation will be off/wrong.
// TODO(marpan): Update these metrics to account for multiple key frames
// in the stream.
82 83
static void set_rate_control_metrics(struct RateControlMetrics *rc,
                                     vpx_codec_enc_cfg_t *cfg) {
84
  unsigned int i = 0;
85 86
  // Set the layer (cumulative) framerate and the target layer (non-cumulative)
  // 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 91 92 93 94
      rc->layer_framerate[0];
  for (i = 0; i < cfg->ts_number_layers; ++i) {
    if (i > 0) {
      rc->layer_framerate[i] = framerate / cfg->ts_rate_decimator[i];
      rc->layer_pfb[i] = 1000.0 *
95
          (rc->layer_target_bitrate[i] - rc->layer_target_bitrate[i - 1]) /
96 97 98 99 100 101 102 103 104
          (rc->layer_framerate[i] - rc->layer_framerate[i - 1]);
    }
    rc->layer_input_frames[i] = 0;
    rc->layer_enc_frames[i] = 0;
    rc->layer_tot_enc_frames[i] = 0;
    rc->layer_encoding_bitrate[i] = 0.0;
    rc->layer_avg_frame_size[i] = 0.0;
    rc->layer_avg_rate_mismatch[i] = 0.0;
  }
105 106 107 108
  rc->window_count = 0;
  rc->window_size = 15;
  rc->avg_st_encoding_bitrate = 0.0;
  rc->variance_st_encoding_bitrate = 0.0;
109 110 111 112 113
}

static void printout_rate_control_summary(struct RateControlMetrics *rc,
                                          vpx_codec_enc_cfg_t *cfg,
                                          int frame_cnt) {
114
  unsigned int i = 0;
115
  int tot_num_frames = 0;
116
  double perc_fluctuation = 0.0;
117 118 119 120 121 122 123
  printf("Total number of processed frames: %d\n\n", frame_cnt -1);
  printf("Rate control layer stats for %d layer(s):\n\n",
      cfg->ts_number_layers);
  for (i = 0; i < cfg->ts_number_layers; ++i) {
    const int num_dropped = (i > 0) ?
        (rc->layer_input_frames[i] - rc->layer_enc_frames[i]) :
        (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;
127 128 129 130 131
    rc->layer_avg_frame_size[i] = rc->layer_avg_frame_size[i] /
        rc->layer_enc_frames[i];
    rc->layer_avg_rate_mismatch[i] = 100.0 * rc->layer_avg_rate_mismatch[i] /
        rc->layer_enc_frames[i];
    printf("For layer#: %d \n", i);
132
    printf("Bitrate (target vs actual): %d %f \n", rc->layer_target_bitrate[i],
133 134 135 136 137 138 139 140 141 142
           rc->layer_encoding_bitrate[i]);
    printf("Average frame size (target vs actual): %f %f \n", rc->layer_pfb[i],
           rc->layer_avg_frame_size[i]);
    printf("Average rate_mismatch: %f \n", rc->layer_avg_rate_mismatch[i]);
    printf("Number of input frames, encoded (non-key) frames, "
        "and perc dropped frames: %d %d %f \n", rc->layer_input_frames[i],
        rc->layer_enc_frames[i],
        100.0 * num_dropped / rc->layer_input_frames[i]);
    printf("\n");
  }
143 144 145 146 147 148 149 150 151 152 153
  rc->avg_st_encoding_bitrate = rc->avg_st_encoding_bitrate / rc->window_count;
  rc->variance_st_encoding_bitrate =
      rc->variance_st_encoding_bitrate / rc->window_count -
      (rc->avg_st_encoding_bitrate * rc->avg_st_encoding_bitrate);
  perc_fluctuation = 100.0 * sqrt(rc->variance_st_encoding_bitrate) /
      rc->avg_st_encoding_bitrate;
  printf("Short-time stats, for window of %d frames: \n",rc->window_size);
  printf("Average, rms-variance, and percent-fluct: %f %f %f \n",
         rc->avg_st_encoding_bitrate,
         sqrt(rc->variance_st_encoding_bitrate),
         perc_fluctuation);
154
  if ((frame_cnt - 1) != tot_num_frames)
155 156 157
    die("Error: Number of input frames not equal to output! \n");
}

158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466
// Temporal scaling parameters:
// NOTE: The 3 prediction frames cannot be used interchangeably due to
// differences in the way they are handled throughout the code. The
// frames should be allocated to layers in the order LAST, GF, ARF.
// Other combinations work, but may produce slightly inferior results.
static void set_temporal_layer_pattern(int layering_mode,
                                       vpx_codec_enc_cfg_t *cfg,
                                       int *layer_flags,
                                       int *flag_periodicity) {
  switch (layering_mode) {
    case 0: {
      // 1-layer.
      int ids[1] = {0};
      cfg->ts_periodicity = 1;
      *flag_periodicity = 1;
      cfg->ts_number_layers = 1;
      cfg->ts_rate_decimator[0] = 1;
      memcpy(cfg->ts_layer_id, ids, sizeof(ids));
      // Update L only.
      layer_flags[0] = VPX_EFLAG_FORCE_KF  | VP8_EFLAG_NO_UPD_GF |
          VP8_EFLAG_NO_UPD_ARF;
      break;
    }
    case 1: {
      // 2-layers, 2-frame period.
      int ids[2] = {0, 1};
      cfg->ts_periodicity = 2;
      *flag_periodicity = 2;
      cfg->ts_number_layers = 2;
      cfg->ts_rate_decimator[0] = 2;
      cfg->ts_rate_decimator[1] = 1;
      memcpy(cfg->ts_layer_id, ids, sizeof(ids));
#if 1
      // 0=L, 1=GF, Intra-layer prediction enabled.
      layer_flags[0] = VPX_EFLAG_FORCE_KF  | VP8_EFLAG_NO_UPD_GF |
          VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF;
      layer_flags[1] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST |
          VP8_EFLAG_NO_REF_ARF;
#else
       // 0=L, 1=GF, Intra-layer prediction disabled.
      layer_flags[0] = VPX_EFLAG_FORCE_KF  | VP8_EFLAG_NO_UPD_GF |
          VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF;
      layer_flags[1] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST |
          VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_REF_LAST;
#endif
      break;
    }
    case 2: {
      // 2-layers, 3-frame period.
      int ids[3] = {0, 1, 1};
      cfg->ts_periodicity = 3;
      *flag_periodicity = 3;
      cfg->ts_number_layers = 2;
      cfg->ts_rate_decimator[0] = 3;
      cfg->ts_rate_decimator[1] = 1;
      memcpy(cfg->ts_layer_id, ids, sizeof(ids));
      // 0=L, 1=GF, Intra-layer prediction enabled.
      layer_flags[0] = VPX_EFLAG_FORCE_KF  | VP8_EFLAG_NO_REF_GF |
          VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
      layer_flags[1] =
      layer_flags[2] = VP8_EFLAG_NO_REF_GF  | VP8_EFLAG_NO_REF_ARF |
          VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST;
      break;
    }
    case 3: {
      // 3-layers, 6-frame period.
      int ids[6] = {0, 2, 2, 1, 2, 2};
      cfg->ts_periodicity = 6;
      *flag_periodicity = 6;
      cfg->ts_number_layers = 3;
      cfg->ts_rate_decimator[0] = 6;
      cfg->ts_rate_decimator[1] = 3;
      cfg->ts_rate_decimator[2] = 1;
      memcpy(cfg->ts_layer_id, ids, sizeof(ids));
      // 0=L, 1=GF, 2=ARF, Intra-layer prediction enabled.
      layer_flags[0] = VPX_EFLAG_FORCE_KF  | VP8_EFLAG_NO_REF_GF |
          VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
      layer_flags[3] = VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_ARF |
          VP8_EFLAG_NO_UPD_LAST;
      layer_flags[1] =
      layer_flags[2] =
      layer_flags[4] =
      layer_flags[5] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_LAST;
      break;
    }
    case 4: {
      // 3-layers, 4-frame period.
      int ids[4] = {0, 2, 1, 2};
      cfg->ts_periodicity = 4;
      *flag_periodicity = 4;
      cfg->ts_number_layers = 3;
      cfg->ts_rate_decimator[0] = 4;
      cfg->ts_rate_decimator[1] = 2;
      cfg->ts_rate_decimator[2] = 1;
      memcpy(cfg->ts_layer_id, ids, sizeof(ids));
      // 0=L, 1=GF, 2=ARF, Intra-layer prediction disabled.
      layer_flags[0] = VPX_EFLAG_FORCE_KF  | VP8_EFLAG_NO_REF_GF |
          VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
      layer_flags[2] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
          VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST;
      layer_flags[1] =
      layer_flags[3] = VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_LAST |
          VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
      break;
    }
    case 5: {
      // 3-layers, 4-frame period.
      int ids[4] = {0, 2, 1, 2};
      cfg->ts_periodicity = 4;
      *flag_periodicity = 4;
      cfg->ts_number_layers     = 3;
      cfg->ts_rate_decimator[0] = 4;
      cfg->ts_rate_decimator[1] = 2;
      cfg->ts_rate_decimator[2] = 1;
      memcpy(cfg->ts_layer_id, ids, sizeof(ids));
      // 0=L, 1=GF, 2=ARF, Intra-layer prediction enabled in layer 1, disabled
      // in layer 2.
      layer_flags[0] = VPX_EFLAG_FORCE_KF  | VP8_EFLAG_NO_REF_GF |
          VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
      layer_flags[2] = VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_LAST |
          VP8_EFLAG_NO_UPD_ARF;
      layer_flags[1] =
      layer_flags[3] = VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_LAST |
          VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
      break;
    }
    case 6: {
      // 3-layers, 4-frame period.
      int ids[4] = {0, 2, 1, 2};
      cfg->ts_periodicity = 4;
      *flag_periodicity = 4;
      cfg->ts_number_layers = 3;
      cfg->ts_rate_decimator[0] = 4;
      cfg->ts_rate_decimator[1] = 2;
      cfg->ts_rate_decimator[2] = 1;
      memcpy(cfg->ts_layer_id, ids, sizeof(ids));
      // 0=L, 1=GF, 2=ARF, Intra-layer prediction enabled.
      layer_flags[0] = VPX_EFLAG_FORCE_KF  | VP8_EFLAG_NO_REF_GF |
          VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
      layer_flags[2] = VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_LAST |
          VP8_EFLAG_NO_UPD_ARF;
      layer_flags[1] =
      layer_flags[3] = VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF;
      break;
    }
    case 7: {
      // NOTE: Probably of academic interest only.
      // 5-layers, 16-frame period.
      int ids[16] = {0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4};
      cfg->ts_periodicity = 16;
      *flag_periodicity = 16;
      cfg->ts_number_layers = 5;
      cfg->ts_rate_decimator[0] = 16;
      cfg->ts_rate_decimator[1] = 8;
      cfg->ts_rate_decimator[2] = 4;
      cfg->ts_rate_decimator[3] = 2;
      cfg->ts_rate_decimator[4] = 1;
      memcpy(cfg->ts_layer_id, ids, sizeof(ids));
      layer_flags[0]  = VPX_EFLAG_FORCE_KF;
      layer_flags[1]  =
      layer_flags[3]  =
      layer_flags[5]  =
      layer_flags[7]  =
      layer_flags[9]  =
      layer_flags[11] =
      layer_flags[13] =
      layer_flags[15] = VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF |
          VP8_EFLAG_NO_UPD_ARF;
      layer_flags[2]  =
      layer_flags[6]  =
      layer_flags[10] =
      layer_flags[14] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_GF;
      layer_flags[4] =
      layer_flags[12] = VP8_EFLAG_NO_REF_LAST | VP8_EFLAG_NO_UPD_ARF;
      layer_flags[8]  = VP8_EFLAG_NO_REF_LAST | VP8_EFLAG_NO_REF_GF;
      break;
    }
    case 8: {
      // 2-layers, with sync point at first frame of layer 1.
      int ids[2] = {0, 1};
      cfg->ts_periodicity = 2;
      *flag_periodicity = 8;
      cfg->ts_number_layers = 2;
      cfg->ts_rate_decimator[0] = 2;
      cfg->ts_rate_decimator[1] = 1;
      memcpy(cfg->ts_layer_id, ids, sizeof(ids));
      // 0=L, 1=GF.
      // ARF is used as predictor for all frames, and is only updated on
      // key frame. Sync point every 8 frames.

      // Layer 0: predict from L and ARF, update L and G.
      layer_flags[0] = VPX_EFLAG_FORCE_KF  | VP8_EFLAG_NO_REF_GF |
          VP8_EFLAG_NO_UPD_ARF;
      // Layer 1: sync point: predict from L and ARF, and update G.
      layer_flags[1] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_UPD_LAST |
          VP8_EFLAG_NO_UPD_ARF;
      // Layer 0, predict from L and ARF, update L.
      layer_flags[2] = VP8_EFLAG_NO_REF_GF  | VP8_EFLAG_NO_UPD_GF |
          VP8_EFLAG_NO_UPD_ARF;
      // Layer 1: predict from L, G and ARF, and update G.
      layer_flags[3] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST |
          VP8_EFLAG_NO_UPD_ENTROPY;
      // Layer 0.
      layer_flags[4] = layer_flags[2];
      // Layer 1.
      layer_flags[5] = layer_flags[3];
      // Layer 0.
      layer_flags[6] = layer_flags[4];
      // Layer 1.
      layer_flags[7] = layer_flags[5];
     break;
    }
    case 9: {
      // 3-layers: Sync points for layer 1 and 2 every 8 frames.
      int ids[4] = {0, 2, 1, 2};
      cfg->ts_periodicity = 4;
      *flag_periodicity = 8;
      cfg->ts_number_layers = 3;
      cfg->ts_rate_decimator[0] = 4;
      cfg->ts_rate_decimator[1] = 2;
      cfg->ts_rate_decimator[2] = 1;
      memcpy(cfg->ts_layer_id, ids, sizeof(ids));
      // 0=L, 1=GF, 2=ARF.
      layer_flags[0] = VPX_EFLAG_FORCE_KF  | VP8_EFLAG_NO_REF_GF |
          VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
      layer_flags[1] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
          VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF;
      layer_flags[2] = VP8_EFLAG_NO_REF_GF   | VP8_EFLAG_NO_REF_ARF |
          VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_ARF;
      layer_flags[3] =
      layer_flags[5] = VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF;
      layer_flags[4] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF |
          VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF;
      layer_flags[6] = VP8_EFLAG_NO_REF_ARF | VP8_EFLAG_NO_UPD_LAST |
          VP8_EFLAG_NO_UPD_ARF;
      layer_flags[7] = VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF |
          VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_ENTROPY;
      break;
    }
    case 10: {
      // 3-layers structure where ARF is used as predictor for all frames,
      // and is only updated on key frame.
      // Sync points for layer 1 and 2 every 8 frames.

      int ids[4] = {0, 2, 1, 2};
      cfg->ts_periodicity = 4;
      *flag_periodicity = 8;
      cfg->ts_number_layers = 3;
      cfg->ts_rate_decimator[0] = 4;
      cfg->ts_rate_decimator[1] = 2;
      cfg->ts_rate_decimator[2] = 1;
      memcpy(cfg->ts_layer_id, ids, sizeof(ids));
      // 0=L, 1=GF, 2=ARF.
      // Layer 0: predict from L and ARF; update L and G.
      layer_flags[0] = VPX_EFLAG_FORCE_KF | VP8_EFLAG_NO_UPD_ARF |
          VP8_EFLAG_NO_REF_GF;
      // Layer 2: sync point: predict from L and ARF; update none.
      layer_flags[1] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_UPD_GF |
          VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST |
          VP8_EFLAG_NO_UPD_ENTROPY;
      // Layer 1: sync point: predict from L and ARF; update G.
      layer_flags[2] = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_UPD_ARF |
          VP8_EFLAG_NO_UPD_LAST;
      // Layer 2: predict from L, G, ARF; update none.
      layer_flags[3] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
          VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_ENTROPY;
      // Layer 0: predict from L and ARF; update L.
      layer_flags[4] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
          VP8_EFLAG_NO_REF_GF;
      // Layer 2: predict from L, G, ARF; update none.
      layer_flags[5] = layer_flags[3];
      // Layer 1: predict from L, G, ARF; update G.
      layer_flags[6] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST;
      // Layer 2: predict from L, G, ARF; update none.
      layer_flags[7] = layer_flags[3];
      break;
    }
    case 11:
    default: {
      // 3-layers structure as in case 10, but no sync/refresh points for
      // layer 1 and 2.
      int ids[4] = {0, 2, 1, 2};
      cfg->ts_periodicity = 4;
      *flag_periodicity = 8;
      cfg->ts_number_layers = 3;
      cfg->ts_rate_decimator[0] = 4;
      cfg->ts_rate_decimator[1] = 2;
      cfg->ts_rate_decimator[2] = 1;
      memcpy(cfg->ts_layer_id, ids, sizeof(ids));
      // 0=L, 1=GF, 2=ARF.
      // Layer 0: predict from L and ARF; update L.
      layer_flags[0] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
          VP8_EFLAG_NO_REF_GF;
      layer_flags[4] = layer_flags[0];
      // Layer 1: predict from L, G, ARF; update G.
      layer_flags[2] = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST;
      layer_flags[6] = layer_flags[2];
      // Layer 2: predict from L, G, ARF; update none.
      layer_flags[1] = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
          VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_ENTROPY;
      layer_flags[3] = layer_flags[1];
      layer_flags[5] = layer_flags[1];
      layer_flags[7] = layer_flags[1];
      break;
    }
  }
}

int main(int argc, char **argv) {
467
  VpxVideoWriter *outfile[VPX_TS_MAX_LAYERS] = {NULL};
468 469 470 471 472 473 474
  vpx_codec_ctx_t codec;
  vpx_codec_enc_cfg_t cfg;
  int frame_cnt = 0;
  vpx_image_t raw;
  vpx_codec_err_t res;
  unsigned int width;
  unsigned int height;
475
  int speed;
476 477 478
  int frame_avail;
  int got_data;
  int flags = 0;
479
  unsigned int i;
480 481 482 483 484
  int pts = 0;  // PTS starts at 0.
  int frame_duration = 1;  // 1 timebase tick per frame.
  int layering_mode = 0;
  int layer_flags[VPX_TS_MAX_PERIODICITY] = {0};
  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};
487 488 489
#else
  vpx_svc_layer_id_t layer_id = {0};
#endif
490
  const VpxInterface *encoder = NULL;
491
  FILE *infile = NULL;
492
  struct RateControlMetrics rc;
493
  int64_t cx_time = 0;
494 495 496 497 498 499 500 501
  const int min_args_base = 11;
#if CONFIG_VP9_HIGHBITDEPTH
  vpx_bit_depth_t bit_depth = VPX_BITS_8;
  int input_bit_depth = 8;
  const int min_args = min_args_base + 1;
#else
  const int min_args = min_args_base;
#endif  // CONFIG_VP9_HIGHBITDEPTH
502 503 504
  double sum_bitrate = 0.0;
  double sum_bitrate2 = 0.0;
  double framerate  = 30.0;
505 506 507

  exec_name = argv[0];
  // Check usage and arguments.
508 509 510 511 512 513
  if (argc < min_args) {
#if CONFIG_VP9_HIGHBITDEPTH
    die("Usage: %s <infile> <outfile> <codec_type(vp8/vp9)> <width> <height> "
        "<rate_num> <rate_den> <speed> <frame_drop_threshold> <mode> "
        "<Rate_0> ... <Rate_nlayers-1> <bit-depth> \n", argv[0]);
#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
  }

520 521 522 523
  encoder = get_vpx_encoder_by_name(argv[3]);
  if (!encoder)
    die("Unsupported codec.");

524
  printf("Using %s\n", vpx_codec_iface_name(encoder->codec_interface()));
525 526 527 528 529 530 531

  width = strtol(argv[4], NULL, 0);
  height = strtol(argv[5], NULL, 0);
  if (width < 16 || width % 2 || height < 16 || height % 2) {
    die("Invalid resolution: %d x %d", width, height);
  }

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]) {
538 539 540
    die("Invalid number of arguments");
  }

541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564
#if CONFIG_VP9_HIGHBITDEPTH
  switch (strtol(argv[argc-1], NULL, 0)) {
    case 8:
      bit_depth = VPX_BITS_8;
      input_bit_depth = 8;
      break;
    case 10:
      bit_depth = VPX_BITS_10;
      input_bit_depth = 10;
      break;
    case 12:
      bit_depth = VPX_BITS_12;
      input_bit_depth = 12;
      break;
    default:
      die("Invalid bit depth (8, 10, 12) %s", argv[argc-1]);
  }
  if (!vpx_img_alloc(&raw,
                     bit_depth == VPX_BITS_8 ? VPX_IMG_FMT_I420 :
                                               VPX_IMG_FMT_I42016,
                     width, height, 32)) {
    die("Failed to allocate image", width, height);
  }
#else
565 566 567
  if (!vpx_img_alloc(&raw, VPX_IMG_FMT_I420, width, height, 32)) {
    die("Failed to allocate image", width, height);
  }
568
#endif  // CONFIG_VP9_HIGHBITDEPTH
569 570

  // Populate encoder configuration.
571
  res = vpx_codec_enc_config_default(encoder->codec_interface(), &cfg, 0);
572 573 574 575 576 577 578 579 580
  if (res) {
    printf("Failed to get config: %s\n", vpx_codec_err_to_string(res));
    return EXIT_FAILURE;
  }

  // Update the default configuration with our settings.
  cfg.g_w = width;
  cfg.g_h = height;

581 582 583 584 585 586 587 588
#if CONFIG_VP9_HIGHBITDEPTH
  if (bit_depth != VPX_BITS_8) {
    cfg.g_bit_depth = bit_depth;
    cfg.g_input_bit_depth = input_bit_depth;
    cfg.g_profile = 2;
  }
#endif  // CONFIG_VP9_HIGHBITDEPTH

589 590 591 592
  // Timebase format e.g. 30fps: numerator=1, demoninator = 30.
  cfg.g_timebase.num = strtol(argv[6], NULL, 0);
  cfg.g_timebase.den = strtol(argv[7], NULL, 0);

593 594 595 596 597
  speed = strtol(argv[8], NULL, 0);
  if (speed < 0) {
    die("Invalid speed setting: must be positive");
  }

598 599 600
  for (i = min_args_base;
       (int)i < min_args_base + mode_to_num_layers[layering_mode];
       ++i) {
601 602 603 604 605
    rc.layer_target_bitrate[i - 11] = strtol(argv[i], NULL, 0);
    if (strncmp(encoder->name, "vp8", 3) == 0)
      cfg.ts_target_bitrate[i - 11] = rc.layer_target_bitrate[i - 11];
    else if (strncmp(encoder->name, "vp9", 3) == 0)
      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);
610 611 612 613
  cfg.rc_end_usage = VPX_CBR;
  cfg.rc_resize_allowed = 0;
  cfg.rc_min_quantizer = 2;
  cfg.rc_max_quantizer = 56;
614 615
  if (strncmp(encoder->name, "vp9", 3) == 0)
    cfg.rc_max_quantizer = 52;
616 617
  cfg.rc_undershoot_pct = 50;
  cfg.rc_overshoot_pct = 50;
618 619 620 621
  cfg.rc_buf_initial_sz = 500;
  cfg.rc_buf_optimal_sz = 600;
  cfg.rc_buf_sz = 1000;

622 623 624
  // Use 1 thread as default.
  cfg.g_threads = 1;

625 626 627
  // Enable error resilient mode.
  cfg.g_error_resilient = 1;
  cfg.g_lag_in_frames   = 0;
628
  cfg.kf_mode = VPX_KF_AUTO;
629 630 631 632

  // Disable automatic keyframe placement.
  cfg.kf_min_dist = cfg.kf_max_dist = 3000;

633 634
  cfg.temporal_layering_mode = VP9E_TEMPORAL_LAYERING_MODE_BYPASS;

635 636 637 638 639
  set_temporal_layer_pattern(layering_mode,
                             &cfg,
                             layer_flags,
                             &flag_periodicity);

640 641
  set_rate_control_metrics(&rc, &cfg);

642
  // Target bandwidth for the whole stream.
643
  // Set to layer_target_bitrate for highest layer (total bitrate).
644
  cfg.rc_target_bitrate = rc.layer_target_bitrate[cfg.ts_number_layers - 1];
645

646
  // Open input file.
647
  if (!(infile = fopen(argv[1], "rb"))) {
648 649 650
    die("Failed to open %s for reading", argv[1]);
  }

651
  framerate = cfg.g_timebase.den / cfg.g_timebase.num;
652 653
  // Open an output file for each stream.
  for (i = 0; i < cfg.ts_number_layers; ++i) {
654 655
    char file_name[PATH_MAX];
    VpxVideoInfo info;
656
    info.codec_fourcc = encoder->fourcc;
657 658 659 660 661
    info.frame_width = cfg.g_w;
    info.frame_height = cfg.g_h;
    info.time_base.numerator = cfg.g_timebase.num;
    info.time_base.denominator = cfg.g_timebase.den;

662
    snprintf(file_name, sizeof(file_name), "%s_%d.ivf", argv[2], i);
663 664
    outfile[i] = vpx_video_writer_open(file_name, kContainerIVF, &info);
    if (!outfile[i])
665
      die("Failed to open %s for writing", file_name);
666 667

    assert(outfile[i] != NULL);
668 669 670 671 672
  }
  // No spatial layers in this encoder.
  cfg.ss_number_layers = 1;

  // Initialize codec.
673 674 675 676 677
#if CONFIG_VP9_HIGHBITDEPTH
  if (vpx_codec_enc_init(
          &codec, encoder->codec_interface(), &cfg,
          bit_depth == VPX_BITS_8 ? 0 : VPX_CODEC_USE_HIGHBITDEPTH))
#else
678
  if (vpx_codec_enc_init(&codec, encoder->codec_interface(), &cfg, 0))
679
#endif  // CONFIG_VP9_HIGHBITDEPTH
680 681
    die_codec(&codec, "Failed to initialize encoder");

682 683
  if (strncmp(encoder->name, "vp8", 3) == 0) {
    vpx_codec_control(&codec, VP8E_SET_CPUUSED, -speed);
684
    vpx_codec_control(&codec, VP8E_SET_NOISE_SENSITIVITY, kDenoiserOff);
685
    vpx_codec_control(&codec, VP8E_SET_STATIC_THRESHOLD, 0);
686
  } else if (strncmp(encoder->name, "vp9", 3) == 0) {
687 688 689
#if VPX_ENCODER_ABI_VERSION > (4 + VPX_CODEC_ABI_VERSION)
    vpx_svc_extra_cfg_t svc_params;
#endif
690 691
      vpx_codec_control(&codec, VP8E_SET_CPUUSED, speed);
      vpx_codec_control(&codec, VP9E_SET_AQ_MODE, 3);
692
      vpx_codec_control(&codec, VP9E_SET_FRAME_PERIODIC_BOOST, 0);
693
      vpx_codec_control(&codec, VP9E_SET_NOISE_SENSITIVITY, 0);
694
      vpx_codec_control(&codec, VP8E_SET_STATIC_THRESHOLD, 0);
695
      vpx_codec_control(&codec, VP9E_SET_TILE_COLUMNS, (cfg.g_threads >> 1));
696
      if (vpx_codec_control(&codec, VP9E_SET_SVC, layering_mode > 0 ? 1: 0)) {
697
        die_codec(&codec, "Failed to set SVC");
698 699 700 701 702 703 704 705 706
#if VPX_ENCODER_ABI_VERSION > (4 + VPX_CODEC_ABI_VERSION)
      for (i = 0; i < cfg.ts_number_layers; ++i) {
        svc_params.max_quantizers[i] = cfg.rc_max_quantizer;
        svc_params.min_quantizers[i] = cfg.rc_min_quantizer;
      }
      svc_params.scaling_factor_num[0] = cfg.g_h;
      svc_params.scaling_factor_den[0] = cfg.g_h;
      vpx_codec_control(&codec, VP9E_SET_SVC_PARAMETERS, &svc_params);
#endif
707 708
    }
  }
Marco's avatar
Marco committed
709 710 711
  if (strncmp(encoder->name, "vp8", 3) == 0) {
    vpx_codec_control(&codec, VP8E_SET_SCREEN_CONTENT_MODE, 0);
  }
712
  vpx_codec_control(&codec, VP8E_SET_TOKEN_PARTITIONS, 1);
713 714 715
  // This controls the maximum target size of the key frame.
  // For generating smaller key frames, use a smaller max_intra_size_pct
  // value, like 100 or 200.
716
  {
717
    const int max_intra_size_pct = 900;
718 719 720
    vpx_codec_control(&codec, VP8E_SET_MAX_INTRA_BITRATE_PCT,
                      max_intra_size_pct);
  }
721 722 723

  frame_avail = 1;
  while (frame_avail || got_data) {
724
    struct vpx_usec_timer timer;
725 726
    vpx_codec_iter_t iter = NULL;
    const vpx_codec_cx_pkt_t *pkt;
727
#if VPX_ENCODER_ABI_VERSION > (4 + VPX_CODEC_ABI_VERSION)
728 729
    // Update the temporal layer_id. No spatial layers in this test.
    layer_id.spatial_layer_id = 0;
730
#endif
731 732
    layer_id.temporal_layer_id =
        cfg.ts_layer_id[frame_cnt % cfg.ts_periodicity];
733 734
    if (strncmp(encoder->name, "vp9", 3) == 0) {
      vpx_codec_control(&codec, VP9E_SET_SVC_LAYER_ID, &layer_id);
Marco's avatar
Marco committed
735 736 737
    } else if (strncmp(encoder->name, "vp8", 3) == 0) {
      vpx_codec_control(&codec, VP8E_SET_TEMPORAL_LAYER_ID,
                        layer_id.temporal_layer_id);
738
    }
739
    flags = layer_flags[frame_cnt % flag_periodicity];
740 741
    if (layering_mode == 0)
      flags = 0;
742
    frame_avail = vpx_img_read(&raw, infile);
743 744
    if (frame_avail)
      ++rc.layer_input_frames[layer_id.temporal_layer_id];
745
    vpx_usec_timer_start(&timer);
746 747 748 749
    if (vpx_codec_encode(&codec, frame_avail? &raw : NULL, pts, 1, flags,
        VPX_DL_REALTIME)) {
      die_codec(&codec, "Failed to encode frame");
    }
750 751
    vpx_usec_timer_mark(&timer);
    cx_time += vpx_usec_timer_elapsed(&timer);
752 753 754 755 756 757 758 759 760 761 762
    // Reset KF flag.
    if (layering_mode != 7) {
      layer_flags[0] &= ~VPX_EFLAG_FORCE_KF;
    }
    got_data = 0;
    while ( (pkt = vpx_codec_get_cx_data(&codec, &iter)) ) {
      got_data = 1;
      switch (pkt->kind) {
        case VPX_CODEC_CX_FRAME_PKT:
          for (i = cfg.ts_layer_id[frame_cnt % cfg.ts_periodicity];
              i < cfg.ts_number_layers; ++i) {
763 764
            vpx_video_writer_write_frame(outfile[i], pkt->data.frame.buf,
                                         pkt->data.frame.sz, pts);
765 766 767 768 769 770 771 772 773 774 775
            ++rc.layer_tot_enc_frames[i];
            rc.layer_encoding_bitrate[i] += 8.0 * pkt->data.frame.sz;
            // Keep count of rate control stats per layer (for non-key frames).
            if (i == cfg.ts_layer_id[frame_cnt % cfg.ts_periodicity] &&
                !(pkt->data.frame.flags & VPX_FRAME_IS_KEY)) {
              rc.layer_avg_frame_size[i] += 8.0 * pkt->data.frame.sz;
              rc.layer_avg_rate_mismatch[i] +=
                  fabs(8.0 * pkt->data.frame.sz - rc.layer_pfb[i]) /
                  rc.layer_pfb[i];
              ++rc.layer_enc_frames[i];
            }
776
          }
777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803
          // Update for short-time encoding bitrate states, for moving window
          // of size rc->window, shifted by rc->window / 2.
          // Ignore first window segment, due to key frame.
          if (frame_cnt > rc.window_size) {
            sum_bitrate += 0.001 * 8.0 * pkt->data.frame.sz * framerate;
            if (frame_cnt % rc.window_size == 0) {
              rc.window_count += 1;
              rc.avg_st_encoding_bitrate += sum_bitrate / rc.window_size;
              rc.variance_st_encoding_bitrate +=
                  (sum_bitrate / rc.window_size) *
                  (sum_bitrate / rc.window_size);
              sum_bitrate = 0.0;
            }
          }
          // Second shifted window.
          if (frame_cnt > rc.window_size + rc.window_size / 2) {
            sum_bitrate2 += 0.001 * 8.0 * pkt->data.frame.sz * framerate;
            if (frame_cnt > 2 * rc.window_size &&
                frame_cnt % rc.window_size == 0) {
              rc.window_count += 1;
              rc.avg_st_encoding_bitrate += sum_bitrate2 / rc.window_size;
              rc.variance_st_encoding_bitrate +=
                  (sum_bitrate2 / rc.window_size) *
                  (sum_bitrate2 / rc.window_size);
              sum_bitrate2 = 0.0;
            }
          }
804 805 806 807 808 809 810 811
          break;
          default:
            break;
      }
    }
    ++frame_cnt;
    pts += frame_duration;
  }
812
  fclose(infile);
813
  printout_rate_control_summary(&rc, &cfg, frame_cnt);
814 815 816 817 818
  printf("\n");
  printf("Frame cnt and encoding time/FPS stats for encoding: %d %f %f \n",
          frame_cnt,
          1000 * (float)cx_time / (double)(frame_cnt * 1000000),
          1000000 * (double)frame_cnt / (double)cx_time);
819

820
  if (vpx_codec_destroy(&codec))
821
    die_codec(&codec, "Failed to destroy codec");
822

823
  // Try to rewrite the output file headers with the actual frame count.
824 825 826
  for (i = 0; i < cfg.ts_number_layers; ++i)
    vpx_video_writer_close(outfile[i]);

827
  vpx_img_free(&raw);
828 829
  return EXIT_SUCCESS;
}