error_resilience_test.cc 17.4 KB
Newer Older
1
/*
Frank Galligan's avatar
Frank Galligan committed
2 3 4 5 6 7 8 9
 *  Copyright (c) 2013 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.
 */
10

11
#include "third_party/googletest/src/include/gtest/gtest.h"
12
#include "test/codec_factory.h"
13 14
#include "test/encode_test_driver.h"
#include "test/i420_video_source.h"
15
#include "test/util.h"
16 17 18

namespace {

19 20
const int kMaxErrorFrames = 12;
const int kMaxDroppableFrames = 12;
21

22
class ErrorResilienceTestLarge : public ::libvpx_test::EncoderTest,
23
    public ::libvpx_test::CodecTestWithParam<libvpx_test::TestMode> {
24
 protected:
25 26 27 28 29 30 31
  ErrorResilienceTestLarge()
      : EncoderTest(GET_PARAM(0)),
        psnr_(0.0),
        nframes_(0),
        mismatch_psnr_(0.0),
        mismatch_nframes_(0),
        encoding_mode_(GET_PARAM(1)) {
32 33
    Reset();
  }
34

35
  virtual ~ErrorResilienceTestLarge() {}
36

37 38 39 40 41
  void Reset() {
    error_nframes_ = 0;
    droppable_nframes_ = 0;
  }

42 43 44 45 46 47 48 49
  virtual void SetUp() {
    InitializeConfig();
    SetMode(encoding_mode_);
  }

  virtual void BeginPassHook(unsigned int /*pass*/) {
    psnr_ = 0.0;
    nframes_ = 0;
50 51
    mismatch_psnr_ = 0.0;
    mismatch_nframes_ = 0;
52 53 54 55 56 57 58
  }

  virtual void PSNRPktHook(const vpx_codec_cx_pkt_t *pkt) {
    psnr_ += pkt->data.psnr.psnr[0];
    nframes_++;
  }

59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83
  //
  // Frame flags and layer id for temporal layers.
  // For two layers, test pattern is:
  //   1     3
  // 0    2     .....
  // LAST is updated on base/layer 0, GOLDEN  updated on layer 1.
  int SetFrameFlags(int frame_num, int num_temp_layers) {
    int frame_flags = 0;
    if (num_temp_layers == 2) {
      if (frame_num % 2 == 0) {
        // Layer 0: predict from L and ARF, update L.
        frame_flags = VP8_EFLAG_NO_REF_GF |
                      VP8_EFLAG_NO_UPD_GF |
                      VP8_EFLAG_NO_UPD_ARF;
      } else {
        // Layer 1: predict from L, GF, and ARF, and update GF.
        frame_flags = VP8_EFLAG_NO_UPD_ARF |
                      VP8_EFLAG_NO_UPD_LAST;
      }
    }
    return frame_flags;
  }

  virtual void PreEncodeFrameHook(libvpx_test::VideoSource *video,
                                  ::libvpx_test::Encoder *encoder) {
84 85 86
    frame_flags_ &= ~(VP8_EFLAG_NO_UPD_LAST |
                      VP8_EFLAG_NO_UPD_GF |
                      VP8_EFLAG_NO_UPD_ARF);
87 88 89
    // For temporal layer case.
    if (cfg_.ts_number_layers > 1) {
      frame_flags_ = SetFrameFlags(video->frame(), cfg_.ts_number_layers);
90
      for (unsigned int i = 0; i < droppable_nframes_; ++i) {
91
        if (droppable_frames_[i] == video->frame()) {
92
          std::cout << "Encoding droppable frame: "
93 94 95
                    << droppable_frames_[i] << "\n";
        }
      }
96 97 98 99 100 101 102 103 104 105 106 107 108 109
    } else {
       if (droppable_nframes_ > 0 &&
         (cfg_.g_pass == VPX_RC_LAST_PASS || cfg_.g_pass == VPX_RC_ONE_PASS)) {
         for (unsigned int i = 0; i < droppable_nframes_; ++i) {
           if (droppable_frames_[i] == video->frame()) {
             std::cout << "Encoding droppable frame: "
                       << droppable_frames_[i] << "\n";
             frame_flags_ |= (VP8_EFLAG_NO_UPD_LAST |
                              VP8_EFLAG_NO_UPD_GF |
                              VP8_EFLAG_NO_UPD_ARF);
             return;
           }
         }
       }
110 111 112
    }
  }

113 114 115 116 117 118
  double GetAveragePsnr() const {
    if (nframes_)
      return psnr_ / nframes_;
    return 0.0;
  }

119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170
  double GetAverageMismatchPsnr() const {
    if (mismatch_nframes_)
      return mismatch_psnr_ / mismatch_nframes_;
    return 0.0;
  }

  virtual bool DoDecode() const {
    if (error_nframes_ > 0 &&
        (cfg_.g_pass == VPX_RC_LAST_PASS || cfg_.g_pass == VPX_RC_ONE_PASS)) {
      for (unsigned int i = 0; i < error_nframes_; ++i) {
        if (error_frames_[i] == nframes_ - 1) {
          std::cout << "             Skipping decoding frame: "
                    << error_frames_[i] << "\n";
          return 0;
        }
      }
    }
    return 1;
  }

  virtual void MismatchHook(const vpx_image_t *img1,
                            const vpx_image_t *img2) {
    double mismatch_psnr = compute_psnr(img1, img2);
    mismatch_psnr_ += mismatch_psnr;
    ++mismatch_nframes_;
    // std::cout << "Mismatch frame psnr: " << mismatch_psnr << "\n";
  }

  void SetErrorFrames(int num, unsigned int *list) {
    if (num > kMaxErrorFrames)
      num = kMaxErrorFrames;
    else if (num < 0)
      num = 0;
    error_nframes_ = num;
    for (unsigned int i = 0; i < error_nframes_; ++i)
      error_frames_[i] = list[i];
  }

  void SetDroppableFrames(int num, unsigned int *list) {
    if (num > kMaxDroppableFrames)
      num = kMaxDroppableFrames;
    else if (num < 0)
      num = 0;
    droppable_nframes_ = num;
    for (unsigned int i = 0; i < droppable_nframes_; ++i)
      droppable_frames_[i] = list[i];
  }

  unsigned int GetMismatchFrames() {
    return mismatch_nframes_;
  }

171 172 173
 private:
  double psnr_;
  unsigned int nframes_;
174 175 176 177 178 179
  unsigned int error_nframes_;
  unsigned int droppable_nframes_;
  double mismatch_psnr_;
  unsigned int mismatch_nframes_;
  unsigned int error_frames_[kMaxErrorFrames];
  unsigned int droppable_frames_[kMaxDroppableFrames];
180 181 182
  libvpx_test::TestMode encoding_mode_;
};

183
TEST_P(ErrorResilienceTestLarge, OnVersusOff) {
184 185 186
  const vpx_rational timebase = { 33333333, 1000000000 };
  cfg_.g_timebase = timebase;
  cfg_.rc_target_bitrate = 2000;
187
  cfg_.g_lag_in_frames = 10;
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

  init_flags_ = VPX_CODEC_USE_PSNR;

  libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
                                     timebase.den, timebase.num, 0, 30);

  // Error resilient mode OFF.
  cfg_.g_error_resilient = 0;
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
  const double psnr_resilience_off = GetAveragePsnr();
  EXPECT_GT(psnr_resilience_off, 25.0);

  // Error resilient mode ON.
  cfg_.g_error_resilient = 1;
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
  const double psnr_resilience_on = GetAveragePsnr();
  EXPECT_GT(psnr_resilience_on, 25.0);

  // Test that turning on error resilient mode hurts by 10% at most.
  if (psnr_resilience_off > 0.0) {
    const double psnr_ratio = psnr_resilience_on / psnr_resilience_off;
    EXPECT_GE(psnr_ratio, 0.9);
    EXPECT_LE(psnr_ratio, 1.1);
  }
}

214 215 216 217
// Check for successful decoding and no encoder/decoder mismatch
// if we lose (i.e., drop before decoding) a set of droppable
// frames (i.e., frames that don't update any reference buffers).
// Check both isolated and consecutive loss.
218
TEST_P(ErrorResilienceTestLarge, DropFramesWithoutRecovery) {
219 220
  const vpx_rational timebase = { 33333333, 1000000000 };
  cfg_.g_timebase = timebase;
221
  cfg_.rc_target_bitrate = 500;
222 223 224
  // FIXME(debargha): Fix this to work for any lag.
  // Currently this test only works for lag = 0
  cfg_.g_lag_in_frames = 0;
225 226 227 228

  init_flags_ = VPX_CODEC_USE_PSNR;

  libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
229
                                     timebase.den, timebase.num, 0, 40);
230 231 232

  // Error resilient mode ON.
  cfg_.g_error_resilient = 1;
233 234 235 236 237 238 239 240
  cfg_.kf_mode = VPX_KF_DISABLED;

  // Set an arbitrary set of error frames same as droppable frames.
  // In addition to isolated loss/drop, add a long consecutive series
  // (of size 9) of dropped frames.
  unsigned int num_droppable_frames = 11;
  unsigned int droppable_frame_list[] = {5, 16, 22, 23, 24, 25, 26, 27, 28,
                                         29, 30};
241 242 243 244 245 246 247 248
  SetDroppableFrames(num_droppable_frames, droppable_frame_list);
  SetErrorFrames(num_droppable_frames, droppable_frame_list);
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
  // Test that no mismatches have been found
  std::cout << "             Mismatch frames: "
            << GetMismatchFrames() << "\n";
  EXPECT_EQ(GetMismatchFrames(), (unsigned int) 0);

249
  // Reset previously set of error/droppable frames.
250 251
  Reset();

252 253 254 255 256
#if 0
  // TODO(jkoleszar): This test is disabled for the time being as too
  // sensitive. It's not clear how to set a reasonable threshold for
  // this behavior.

257 258 259 260 261
  // Now set an arbitrary set of error frames that are non-droppable
  unsigned int num_error_frames = 3;
  unsigned int error_frame_list[] = {3, 10, 20};
  SetErrorFrames(num_error_frames, error_frame_list);
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
262

263 264 265 266 267 268 269 270
  // Test that dropping an arbitrary set of inter frames does not hurt too much
  // Note the Average Mismatch PSNR is the average of the PSNR between
  // decoded frame and encoder's version of the same frame for all frames
  // with mismatch.
  const double psnr_resilience_mismatch = GetAverageMismatchPsnr();
  std::cout << "             Mismatch PSNR: "
            << psnr_resilience_mismatch << "\n";
  EXPECT_GT(psnr_resilience_mismatch, 20.0);
271
#endif
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
// Check for successful decoding and no encoder/decoder mismatch
// if we lose (i.e., drop before decoding) the enhancement layer frames for a
// two layer temporal pattern. The base layer does not predict from the top
// layer, so successful decoding is expected.
TEST_P(ErrorResilienceTestLarge, 2LayersDropEnhancement) {
  const vpx_rational timebase = { 33333333, 1000000000 };
  cfg_.g_timebase = timebase;
  cfg_.rc_target_bitrate = 500;
  cfg_.g_lag_in_frames = 0;

  cfg_.rc_end_usage = VPX_CBR;
  // 2 Temporal layers, no spatial layers, CBR mode.
  cfg_.ss_number_layers = 1;
  cfg_.ts_number_layers = 2;
  cfg_.ts_rate_decimator[0] = 2;
  cfg_.ts_rate_decimator[1] = 1;
  cfg_.ts_periodicity = 2;
  cfg_.ts_target_bitrate[0] = 60 * cfg_.rc_target_bitrate / 100;
  cfg_.ts_target_bitrate[1] = cfg_.rc_target_bitrate;

  init_flags_ = VPX_CODEC_USE_PSNR;

  libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
                                     timebase.den, timebase.num, 0, 40);

  // Error resilient mode ON.
  cfg_.g_error_resilient = 1;
  cfg_.kf_mode = VPX_KF_DISABLED;

  // The odd frames are the enhancement layer for 2 layer pattern, so set
  // those frames as droppable. Drop the last 7 frames.
  unsigned int num_droppable_frames = 7;
  unsigned int droppable_frame_list[] = {27, 29, 31, 33, 35, 37, 39};
  SetDroppableFrames(num_droppable_frames, droppable_frame_list);
  SetErrorFrames(num_droppable_frames, droppable_frame_list);
  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
  // Test that no mismatches have been found
  std::cout << "             Mismatch frames: "
            << GetMismatchFrames() << "\n";
  EXPECT_EQ(GetMismatchFrames(), (unsigned int) 0);

  // Reset previously set of error/droppable frames.
  Reset();
}

Marco's avatar
Marco committed
319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515
class ErrorResilienceTestLargeCodecControls : public ::libvpx_test::EncoderTest,
    public ::libvpx_test::CodecTestWithParam<libvpx_test::TestMode> {
 protected:
  ErrorResilienceTestLargeCodecControls()
      : EncoderTest(GET_PARAM(0)),
        encoding_mode_(GET_PARAM(1)) {
    Reset();
  }

  virtual ~ErrorResilienceTestLargeCodecControls() {}

  void Reset() {
    last_pts_ = 0;
    tot_frame_number_ = 0;
    // For testing up to 3 layers.
    for (int i = 0; i < 3; ++i) {
      bits_total_[i] = 0;
    }
    duration_ = 0.0;
  }

  virtual void SetUp() {
    InitializeConfig();
    SetMode(encoding_mode_);
  }

  //
  // Frame flags and layer id for temporal layers.
  //

  // For two layers, test pattern is:
  //   1     3
  // 0    2     .....
  // For three layers, test pattern is:
  //   1      3    5      7
  //      2           6
  // 0          4            ....
  // LAST is always update on base/layer 0, GOLDEN is updated on layer 1,
  // and ALTREF is updated on top layer for 3 layer pattern.
  int SetFrameFlags(int frame_num, int num_temp_layers) {
    int frame_flags = 0;
    if (num_temp_layers == 2) {
      if (frame_num % 2 == 0) {
        // Layer 0: predict from L and ARF, update L.
        frame_flags = VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_UPD_GF |
                      VP8_EFLAG_NO_UPD_ARF;
      } else {
        // Layer 1: predict from L, G and ARF, and update G.
        frame_flags = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST |
                      VP8_EFLAG_NO_UPD_ENTROPY;
      }
    } else if (num_temp_layers == 3) {
      if (frame_num % 4 == 0) {
        // Layer 0: predict from L, update L.
        frame_flags = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_ARF |
                      VP8_EFLAG_NO_REF_GF | VP8_EFLAG_NO_REF_ARF;
      } else if ((frame_num - 2) % 4 == 0) {
        // Layer 1: predict from L, G,  update G.
        frame_flags = VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_NO_UPD_LAST |
                      VP8_EFLAG_NO_REF_ARF;
      }  else if ((frame_num - 1) % 2 == 0) {
        // Layer 2: predict from L, G, ARF; update ARG.
        frame_flags = VP8_EFLAG_NO_UPD_GF | VP8_EFLAG_NO_UPD_LAST;
      }
    }
    return frame_flags;
  }

  int SetLayerId(int frame_num, int num_temp_layers) {
    int layer_id = 0;
    if (num_temp_layers == 2) {
      if (frame_num % 2 == 0) {
        layer_id = 0;
      } else {
         layer_id = 1;
      }
    } else if (num_temp_layers == 3) {
      if (frame_num % 4 == 0) {
        layer_id = 0;
      } else if ((frame_num - 2) % 4 == 0) {
        layer_id = 1;
      } else if ((frame_num - 1) % 2 == 0) {
        layer_id = 2;
      }
    }
    return layer_id;
  }

  virtual void PreEncodeFrameHook(libvpx_test::VideoSource *video,
                                  libvpx_test::Encoder *encoder) {
    if (cfg_.ts_number_layers > 1) {
        int layer_id = SetLayerId(video->frame(), cfg_.ts_number_layers);
        int frame_flags = SetFrameFlags(video->frame(), cfg_.ts_number_layers);
        if (video->frame() > 0) {
          encoder->Control(VP8E_SET_TEMPORAL_LAYER_ID, layer_id);
          encoder->Control(VP8E_SET_FRAME_FLAGS, frame_flags);
        }
       const vpx_rational_t tb = video->timebase();
       timebase_ = static_cast<double>(tb.num) / tb.den;
       duration_ = 0;
       return;
    }
  }

  virtual void FramePktHook(const vpx_codec_cx_pkt_t *pkt) {
    // Time since last timestamp = duration.
    vpx_codec_pts_t duration = pkt->data.frame.pts - last_pts_;
    if (duration > 1) {
      // Update counter for total number of frames (#frames input to encoder).
      // Needed for setting the proper layer_id below.
      tot_frame_number_ += static_cast<int>(duration - 1);
    }
    int layer = SetLayerId(tot_frame_number_, cfg_.ts_number_layers);
    const size_t frame_size_in_bits = pkt->data.frame.sz * 8;
    // Update the total encoded bits. For temporal layers, update the cumulative
    // encoded bits per layer.
    for (int i = layer; i < static_cast<int>(cfg_.ts_number_layers); ++i) {
      bits_total_[i] += frame_size_in_bits;
    }
    // Update the most recent pts.
    last_pts_ = pkt->data.frame.pts;
    ++tot_frame_number_;
  }

  virtual void EndPassHook(void) {
    duration_ = (last_pts_ + 1) * timebase_;
    if (cfg_.ts_number_layers  > 1) {
      for (int layer = 0; layer < static_cast<int>(cfg_.ts_number_layers);
          ++layer) {
        if (bits_total_[layer]) {
          // Effective file datarate:
          effective_datarate_[layer] = (bits_total_[layer] / 1000.0) / duration_;
        }
      }
    }
  }

  double effective_datarate_[3];
   private:
    libvpx_test::TestMode encoding_mode_;
    vpx_codec_pts_t last_pts_;
    double timebase_;
    int64_t bits_total_[3];
    double duration_;
    int tot_frame_number_;
  };

// Check two codec controls used for:
// (1) for setting temporal layer id, and (2) for settings encoder flags.
// This test invokes those controls for each frame, and verifies encoder/decoder
// mismatch and basic rate control response.
// TODO(marpan): Maybe move this test to datarate_test.cc.
TEST_P(ErrorResilienceTestLargeCodecControls, CodecControl3TemporalLayers) {
  cfg_.rc_buf_initial_sz = 500;
  cfg_.rc_buf_optimal_sz = 500;
  cfg_.rc_buf_sz = 1000;
  cfg_.rc_dropframe_thresh = 1;
  cfg_.rc_min_quantizer = 2;
  cfg_.rc_max_quantizer = 56;
  cfg_.rc_end_usage = VPX_CBR;
  cfg_.rc_dropframe_thresh = 1;
  cfg_.g_lag_in_frames = 0;
  cfg_.kf_mode = VPX_KF_DISABLED;
  cfg_.g_error_resilient = 1;

  // 3 Temporal layers. Framerate decimation (4, 2, 1).
  cfg_.ts_number_layers = 3;
  cfg_.ts_rate_decimator[0] = 4;
  cfg_.ts_rate_decimator[1] = 2;
  cfg_.ts_rate_decimator[2] = 1;
  cfg_.ts_periodicity = 4;
  cfg_.ts_layer_id[0] = 0;
  cfg_.ts_layer_id[1] = 2;
  cfg_.ts_layer_id[2] = 1;
  cfg_.ts_layer_id[3] = 2;

  ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
                                       30, 1, 0, 200);
  for (int i = 200; i <= 800; i += 200) {
    cfg_.rc_target_bitrate = i;
    Reset();
    // 40-20-40 bitrate allocation for 3 temporal layers.
    cfg_.ts_target_bitrate[0] = 40 * cfg_.rc_target_bitrate / 100;
    cfg_.ts_target_bitrate[1] = 60 * cfg_.rc_target_bitrate / 100;
    cfg_.ts_target_bitrate[2] = cfg_.rc_target_bitrate;
    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
    for (int j = 0; j < static_cast<int>(cfg_.ts_number_layers); ++j) {
      ASSERT_GE(effective_datarate_[j], cfg_.ts_target_bitrate[j] * 0.75)
          << " The datarate for the file is lower than target by too much, "
              "for layer: " << j;
      ASSERT_LE(effective_datarate_[j], cfg_.ts_target_bitrate[j] * 1.25)
          << " The datarate for the file is greater than target by too much, "
              "for layer: " << j;
    }
  }
}

516
VP8_INSTANTIATE_TEST_CASE(ErrorResilienceTestLarge, ONE_PASS_TEST_MODES);
Marco's avatar
Marco committed
517 518
VP8_INSTANTIATE_TEST_CASE(ErrorResilienceTestLargeCodecControls,
                          ONE_PASS_TEST_MODES);
519
VP9_INSTANTIATE_TEST_CASE(ErrorResilienceTestLarge, ONE_PASS_TEST_MODES);
520
}  // namespace