libvpx: Pull from upstream

source/libvpx/examples/vpx_temporal_svc_encoder.c

 /*
  *  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.
  */
 
@@ skipping 43 matching lines @@
   // Framerate per layer layer (cumulative).
   double layer_framerate[VPX_TS_MAX_LAYERS];
   // Target average frame size per layer (per-frame-bandwidth per layer).
   double layer_pfb[VPX_TS_MAX_LAYERS];
   // Actual average frame size per layer.
   double layer_avg_frame_size[VPX_TS_MAX_LAYERS];
   // Average rate mismatch per layer (|target - actual| / target).
   double layer_avg_rate_mismatch[VPX_TS_MAX_LAYERS];
   // Actual encoding bitrate per layer (cumulative).
   double layer_encoding_bitrate[VPX_TS_MAX_LAYERS];
+  // 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;
 };
 
 // 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.
 static void set_rate_control_metrics(struct RateControlMetrics *rc,
                                      vpx_codec_enc_cfg_t *cfg) {
@@ skipping 11 matching lines @@
           (cfg->ts_target_bitrate[i] - cfg->ts_target_bitrate[i - 1]) /
           (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;
   }
+  rc->window_count = 0;
+  rc->window_size = 15;
+  rc->avg_st_encoding_bitrate = 0.0;
+  rc->variance_st_encoding_bitrate = 0.0;
 }
 
 static void printout_rate_control_summary(struct RateControlMetrics *rc,
                                           vpx_codec_enc_cfg_t *cfg,
                                           int frame_cnt) {
   unsigned int i = 0;
   int tot_num_frames = 0;
+  double perc_fluctuation = 0.0;
   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);
     tot_num_frames += rc->layer_input_frames[i];
     rc->layer_encoding_bitrate[i] = 0.001 * rc->layer_framerate[i] *
         rc->layer_encoding_bitrate[i] / tot_num_frames;
     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);
     printf("Bitrate (target vs actual): %d %f \n", cfg->ts_target_bitrate[i],
            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");
   }
+  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);
   if ((frame_cnt - 1) != tot_num_frames)
     die("Error: Number of input frames not equal to output! \n");
 }
 
 // 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,
@@ skipping 324 matching lines @@
   struct RateControlMetrics rc;
   int64_t cx_time = 0;
   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
+  double sum_bitrate = 0.0;
+  double sum_bitrate2 = 0.0;
+  double framerate  = 30.0;
 
   exec_name = argv[0];
   // Check usage and arguments.
   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
     die("Usage: %s <infile> <outfile> <codec_type(vp8/vp9)> <width> <height> "
@@ skipping 115 matching lines @@
 
   // Target bandwidth for the whole stream.
   // Set to ts_target_bitrate for highest layer (total bitrate).
   cfg.rc_target_bitrate = cfg.ts_target_bitrate[cfg.ts_number_layers - 1];
 
   // Open input file.
   if (!(infile = fopen(argv[1], "rb"))) {
     die("Failed to open %s for reading", argv[1]);
   }
 
+  framerate = cfg.g_timebase.den / cfg.g_timebase.num;
   // Open an output file for each stream.
   for (i = 0; i < cfg.ts_number_layers; ++i) {
     char file_name[PATH_MAX];
     VpxVideoInfo info;
     info.codec_fourcc = encoder->fourcc;
     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;
 
@@ skipping 28 matching lines @@
       if (vpx_codec_control(&codec, VP9E_SET_SVC, 1)) {
         die_codec(&codec, "Failed to set SVC");
     }
   }
   vpx_codec_control(&codec, VP8E_SET_STATIC_THRESHOLD, 1);
   vpx_codec_control(&codec, VP8E_SET_TOKEN_PARTITIONS, 1);
   // 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.
   {
-    const int max_intra_size_pct = 200;
+    const int max_intra_size_pct = 900;
     vpx_codec_control(&codec, VP8E_SET_MAX_INTRA_BITRATE_PCT,
                       max_intra_size_pct);
   }
 
   frame_avail = 1;
   while (frame_avail || got_data) {
     struct vpx_usec_timer timer;
     vpx_codec_iter_t iter = NULL;
     const vpx_codec_cx_pkt_t *pkt;
     // Update the temporal layer_id. No spatial layers in this test.
@@ skipping 32 matching lines @@
             // 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];
             }
           }
+          // 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;
+            }
+          }
           break;
           default:
             break;
       }
     }
     ++frame_cnt;
     pts += frame_duration;
   }
   fclose(infile);
   printout_rate_control_summary(&rc, &cfg, frame_cnt);
   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);
 
   if (vpx_codec_destroy(&codec))
     die_codec(&codec, "Failed to destroy codec");
 
   // Try to rewrite the output file headers with the actual frame count.
   for (i = 0; i < cfg.ts_number_layers; ++i)
     vpx_video_writer_close(outfile[i]);
 
   vpx_img_free(&raw);
   return EXIT_SUCCESS;
 }