fix: Parse one frame mpeg-ts video (shaka-project#1015)

Closes shaka-project#1013 Co-authored-by: Joey Parrish <joeyparrish@users.noreply.github.com>
joeyparrish · Oct 28, 2022 · b221aa9 · b221aa9
1 parent ab8ab12
commit b221aa9
Show file tree

Hide file tree

Showing 13 changed files with 141 additions and 28 deletions.
diff --git a/packager/media/codecs/h264_parser.cc b/packager/media/codecs/h264_parser.cc
@@ -112,6 +112,17 @@ bool H264SliceHeader::IsSISlice() const {
     }                                                                      \
   } while (0)
 
+#define READ_LONG_OR_RETURN(out)        \
+  do {                                  \
+    long _out;                          \
+    int _tmp_out;                       \
+    READ_BITS_OR_RETURN(16, &_tmp_out); \
+    _out = (long)(_tmp_out) << 16;      \
+    READ_BITS_OR_RETURN(16, &_tmp_out); \
+    _out |= _tmp_out;                   \
+    *(out) = _out;                      \
+  } while(0)
+
 #define READ_BOOL_OR_RETURN(out)                                           \
   do {                                                                     \
     int _out;                                                              \
@@ -524,14 +535,12 @@ H264Parser::Result H264Parser::ParseVUIParameters(H26xBitReader* br,
     READ_UE_OR_RETURN(&data);  // chroma_sample_loc_type_bottom_field
   }
 
-  // Read and ignore timing info.
-  READ_BOOL_OR_RETURN(&data);  // timing_info_present_flag
-  if (data) {
-    READ_BITS_OR_RETURN(16, &data);  // num_units_in_tick
-    READ_BITS_OR_RETURN(16, &data);  // num_units_in_tick
-    READ_BITS_OR_RETURN(16, &data);  // time_scale
-    READ_BITS_OR_RETURN(16, &data);  // time_scale
-    READ_BOOL_OR_RETURN(&data);  // fixed_frame_rate_flag
+  // Read timing info.
+  READ_BOOL_OR_RETURN(&sps->timing_info_present_flag);
+  if (sps->timing_info_present_flag) {
+    READ_LONG_OR_RETURN(&sps->num_units_in_tick);
+    READ_LONG_OR_RETURN(&sps->time_scale);
+    READ_BOOL_OR_RETURN(&sps->fixed_frame_rate_flag);
   }
 
   // Read and ignore NAL HRD parameters, if present.

diff --git a/packager/media/codecs/h264_parser.h b/packager/media/codecs/h264_parser.h
@@ -82,6 +82,11 @@ struct H264Sps {
   int sar_height;   // Set to 0 when not specified.
   int transfer_characteristics;
 
+  bool timing_info_present_flag;
+  long num_units_in_tick;
+  long time_scale;
+  bool fixed_frame_rate_flag;
+
   bool bitstream_restriction_flag;
   int max_num_reorder_frames;
   int max_dec_frame_buffering;

diff --git a/packager/media/codecs/h265_parser.cc b/packager/media/codecs/h265_parser.cc
@@ -28,6 +28,22 @@
       return status;     \
   } while (false)
 
+#define READ_LONG_OR_RETURN(out)                                           \
+  do {                                                                     \
+    int _top_half, _bottom_half;                                           \
+    if (!br->ReadBits(16, &_top_half)) {                                   \
+      DVLOG(1)                                                             \
+          << "Error in stream: unexpected EOS while trying to read " #out; \
+      return kInvalidStream;                                               \
+    }                                                                      \
+    if (!br->ReadBits(16, &_bottom_half)) {                                \
+      DVLOG(1)                                                             \
+          << "Error in stream: unexpected EOS while trying to read " #out; \
+      return kInvalidStream;                                               \
+    }                                                                      \
+    *(out) = ((long)_top_half) << 16 | _bottom_half;                       \
+  } while (false)
+
 namespace shaka {
 namespace media {
 
@@ -688,11 +704,10 @@ H265Parser::Result H265Parser::ParseVuiParameters(int max_num_sub_layers_minus1,
     TRUE_OR_RETURN(br->ReadUE(&ignored));  // def_disp_win_bottom_offset
   }
 
-  bool vui_timing_info_present_flag;
-  TRUE_OR_RETURN(br->ReadBool(&vui_timing_info_present_flag));
-  if (vui_timing_info_present_flag) {
-    // vui_num_units_in_tick, vui_time_scale
-    TRUE_OR_RETURN(br->SkipBits(32 + 32));
+  TRUE_OR_RETURN(br->ReadBool(&vui->vui_timing_info_present_flag));
+  if (vui->vui_timing_info_present_flag) {
+    READ_LONG_OR_RETURN(&vui->vui_num_units_in_tick);
+    READ_LONG_OR_RETURN(&vui->vui_time_scale);
 
     bool vui_poc_proportional_to_timing_flag;
     TRUE_OR_RETURN(br->ReadBool(&vui_poc_proportional_to_timing_flag));

diff --git a/packager/media/codecs/h265_parser.h b/packager/media/codecs/h265_parser.h
@@ -52,6 +52,10 @@ struct H265VuiParameters {
   int sar_height = 0;
   int transfer_characteristics = 0;
 
+  bool vui_timing_info_present_flag = false;
+  long vui_num_units_in_tick = 0;
+  long vui_time_scale = 0;
+
   bool bitstream_restriction_flag = false;
   int min_spatial_segmentation_idc = 0;
 

diff --git a/packager/media/formats/mp2t/es_parser_h264.cc b/packager/media/formats/mp2t/es_parser_h264.cc
@@ -184,6 +184,22 @@ bool EsParserH264::UpdateVideoDecoderConfig(int pps_id) {
   return true;
 }
 
+int64_t EsParserH264::CalculateSampleDuration(int pps_id) {
+  auto pps = h264_parser_->GetPps(pps_id);
+  if (pps) {
+    auto sps_id = pps->seq_parameter_set_id;
+    auto sps = h264_parser_->GetSps(sps_id);
+    if (sps && sps->timing_info_present_flag && sps->fixed_frame_rate_flag) {
+      return static_cast<int64_t>(kMpeg2Timescale) * sps->num_units_in_tick *
+             2 / sps->time_scale;
+    }
+  }
+  LOG(WARNING) << "[MPEG-2 TS] PID " << pid()
+               << " Cannot calculate frame rate from SPS.";
+  // Returns arbitrary safe duration
+  return 0.001 * kMpeg2Timescale;  // 1ms.
+}
+
 }  // namespace mp2t
 }  // namespace media
 }  // namespace shaka
diff --git a/packager/media/formats/mp2t/es_parser_h264.h b/packager/media/formats/mp2t/es_parser_h264.h
@@ -39,7 +39,8 @@ class EsParserH264 : public EsParserH26x {
   // Update the video decoder config based on an H264 SPS.
   // Return true if successful.
   bool UpdateVideoDecoderConfig(int sps_id) override;
-
+  // Calculate video sample duration based on SPS data
+  int64_t CalculateSampleDuration(int pps_id) override;
   // Callback to pass the stream configuration.
   NewStreamInfoCB new_stream_info_cb_;
 

diff --git a/packager/media/formats/mp2t/es_parser_h265.cc b/packager/media/formats/mp2t/es_parser_h265.cc
@@ -184,6 +184,24 @@ bool EsParserH265::UpdateVideoDecoderConfig(int pps_id) {
   return true;
 }
 
+int64_t EsParserH265::CalculateSampleDuration(int pps_id) {
+  auto pps = h265_parser_->GetPps(pps_id);
+  if (pps) {
+    auto sps_id = pps->seq_parameter_set_id;
+    auto sps = h265_parser_->GetSps(sps_id);
+    if (sps && sps->vui_parameters_present &&
+        sps->vui_parameters.vui_timing_info_present_flag) {
+      return static_cast<int64_t>(kMpeg2Timescale) *
+             sps->vui_parameters.vui_num_units_in_tick * 2 /
+             sps->vui_parameters.vui_time_scale;
+    }
+  }
+  LOG(WARNING) << "[MPEG-2 TS] PID " << pid()
+               << " Cannot calculate frame rate from SPS.";
+  // Returns arbitrary safe duration
+  return 0.001 * kMpeg2Timescale;  // 1ms.
+}
+
 }  // namespace mp2t
 }  // namespace media
 }  // namespace shaka
diff --git a/packager/media/formats/mp2t/es_parser_h265.h b/packager/media/formats/mp2t/es_parser_h265.h
@@ -42,6 +42,7 @@ class EsParserH265 : public EsParserH26x {
   // Return true if successful.
   bool UpdateVideoDecoderConfig(int sps_id) override;
 
+  int64_t CalculateSampleDuration(int pps_id) override;
   // Callback to pass the stream configuration.
   NewStreamInfoCB new_stream_info_cb_;
 

diff --git a/packager/media/formats/mp2t/es_parser_h26x.cc b/packager/media/formats/mp2t/es_parser_h26x.cc
@@ -99,7 +99,9 @@ bool EsParserH26x::Flush() {
 
   if (pending_sample_) {
     // Flush pending sample.
-    DCHECK(pending_sample_duration_);
+    if (!pending_sample_duration_) {
+      pending_sample_duration_ = CalculateSampleDuration(pending_sample_pps_id_);
+    }
     pending_sample_->set_duration(pending_sample_duration_);
     emit_sample_cb_.Run(std::move(pending_sample_));
   }
@@ -330,7 +332,8 @@ bool EsParserH26x::EmitFrame(int64_t access_unit_pos,
       pending_sample_->set_duration(sample_duration);
 
       const int kArbitraryGapScale = 10;
-      if (sample_duration > kArbitraryGapScale * pending_sample_duration_) {
+      if (pending_sample_duration_ &&
+          sample_duration > kArbitraryGapScale * pending_sample_duration_) {
         LOG(WARNING) << "[MPEG-2 TS] PID " << pid() << " Possible GAP at dts "
                      << pending_sample_->dts() << " with next sample at dts "
                      << media_sample->dts() << " (difference "
@@ -342,6 +345,7 @@ bool EsParserH26x::EmitFrame(int64_t access_unit_pos,
     emit_sample_cb_.Run(std::move(pending_sample_));
   }
   pending_sample_ = media_sample;
+  pending_sample_pps_id_ = pps_id;
 
   return true;
 }

diff --git a/packager/media/formats/mp2t/es_parser_h26x.h b/packager/media/formats/mp2t/es_parser_h26x.h
@@ -96,6 +96,8 @@ class EsParserH26x : public EsParser {
                  int access_unit_size,
                  bool is_key_frame,
                  int pps_id);
+  // Calculates frame duration based on SPS frame data
+  virtual int64_t CalculateSampleDuration(int pps_id) = 0;
 
   // Callback to pass the frames.
   EmitSampleCB emit_sample_cb_;
@@ -127,6 +129,7 @@ class EsParserH26x : public EsParser {
 
   // Frame for which we do not yet have a duration.
   std::shared_ptr<MediaSample> pending_sample_;
+  int pending_sample_pps_id_ = -1;
   int64_t pending_sample_duration_ = 0;
 
   // Indicates whether waiting for first key frame.

diff --git a/packager/media/formats/mp2t/es_parser_h26x_unittest.cc b/packager/media/formats/mp2t/es_parser_h26x_unittest.cc
@@ -120,6 +120,11 @@ class TestableEsParser : public EsParserH26x {
   }
 
  private:
+  int64_t CalculateSampleDuration(int pps_id) override {
+    // Typical 40ms - frame duration with 25 FPS
+    return 0.04 * 90000;
+  }
+
   const int kTestPpsId = 123;
 
   Nalu::CodecType codec_type_;
@@ -179,6 +184,7 @@ class EsParserH26xTest : public testing::Test {
     const std::vector<uint8_t> sample_data(
         sample->data(), sample->data() + sample->data_size());
     EXPECT_EQ(samples_[sample_id], sample_data);
+    media_samples_.push_back(sample);
   }
 
   void NewVideoConfig(std::shared_ptr<StreamInfo> config) {
@@ -187,6 +193,7 @@ class EsParserH26xTest : public testing::Test {
 
  protected:
   std::vector<std::vector<uint8_t>> samples_;
+  std::vector<std::shared_ptr<MediaSample>> media_samples_;
   size_t sample_count_;
   bool has_stream_info_;
 };
@@ -377,6 +384,10 @@ TEST_F(EsParserH26xTest, H264BasicSupport) {
   RunTest(Nalu::kH264, kData, arraysize(kData));
   EXPECT_EQ(3u, sample_count_);
   EXPECT_TRUE(has_stream_info_);
+  EXPECT_EQ(3u, media_samples_.size());
+  for (size_t i = 0; i < media_samples_.size(); i++) {
+    EXPECT_GT(media_samples_[i]->duration(), 0u);
+  }
 }
 
 // This is not compliant to H264 spec, but VLC generates streams like this. See
@@ -453,6 +464,33 @@ TEST_F(EsParserH26xTest, H264DoesNotStartOnRsv) {
   EXPECT_TRUE(has_stream_info_);
 }
 
+TEST_F(EsParserH26xTest, H264ContainsOnlyOneFrame) {
+  const H26xNaluType kData[] = {
+      kSeparator,
+      kH264Aud,
+      kH264Sps,
+      kH264VclKeyFrame,
+  };
+
+  RunTest(Nalu::kH264, kData, arraysize(kData));
+  EXPECT_TRUE(has_stream_info_);
+  EXPECT_EQ(1u, sample_count_);
+  EXPECT_EQ(1u, media_samples_.size());
+  EXPECT_GT(media_samples_[0]->duration(), 0u);
+}
+
+TEST_F(EsParserH26xTest, H265ContainsOnlyOneFrame) {
+  const H26xNaluType kData[] = {
+      kSeparator, kH265Aud, kH265Sps, kH265VclKeyFrame,
+  };
+
+  RunTest(Nalu::kH265, kData, arraysize(kData));
+  EXPECT_TRUE(has_stream_info_);
+  EXPECT_EQ(1u, sample_count_);
+  EXPECT_EQ(1u, media_samples_.size());
+  EXPECT_GT(media_samples_[0]->duration(), 0u);
+}
+
 }  // namespace mp2t
 }  // namespace media
 }  // namespace shaka
diff --git a/packager/media/formats/mp2t/mp2t_media_parser.cc b/packager/media/formats/mp2t/mp2t_media_parser.cc
@@ -97,7 +97,7 @@ bool PidState::PushTsPacket(const TsPacket& ts_packet) {
   // just discard the incoming TS packet.
   if (!enable_)
     return true;
-
+  // TODO(bzd): continuity_counter_ is never set
   int expected_continuity_counter = (continuity_counter_ + 1) % 16;
   if (continuity_counter_ >= 0 &&
       ts_packet.continuity_counter() != expected_continuity_counter) {
@@ -218,10 +218,11 @@ bool Mp2tMediaParser::Parse(const uint8_t* buf, int size) {
       ts_byte_queue_.Pop(1);
       continue;
     }
-    DVLOG(LOG_LEVEL_TS)
-        << "Processing PID=" << ts_packet->pid()
-        << " start_unit=" << ts_packet->payload_unit_start_indicator();
-
+    DVLOG(LOG_LEVEL_TS) << "Processing PID=" << ts_packet->pid()
+                        << " start_unit="
+                        << ts_packet->payload_unit_start_indicator()
+                        << " continuity_counter="
+                        << ts_packet->continuity_counter();
     // Parse the section.
     auto it = pids_.find(ts_packet->pid());
     if (it == pids_.end() &&

diff --git a/packager/media/formats/mp2t/ts_section_pes.cc b/packager/media/formats/mp2t/ts_section_pes.cc
@@ -192,7 +192,7 @@ bool TsSectionPes::ParseInternal(const uint8_t* raw_pes, int raw_pes_size) {
   RCHECK(bit_reader.ReadBits(16, &pes_packet_length));
 
   RCHECK(packet_start_code_prefix == kPesStartCode);
-  DVLOG(LOG_LEVEL_PES) << "stream_id=" << std::hex << stream_id << std::dec;
+  DVLOG(LOG_LEVEL_PES) << "stream_id=" << stream_id;
   if (pes_packet_length == 0)
     pes_packet_length = static_cast<int>(bit_reader.bits_available()) / 8;
 
@@ -303,12 +303,10 @@ bool TsSectionPes::ParseInternal(const uint8_t* raw_pes, int raw_pes_size) {
   RCHECK(pes_header_remaining_size >= 0);
 
   // Read the PES packet.
-  DVLOG(LOG_LEVEL_PES)
-      << "Emit a reassembled PES:"
-      << " size=" << es_size
-      << " pts=" << media_pts
-      << " dts=" << media_dts
-      << " data_alignment_indicator=" << data_alignment_indicator;
+  DVLOG(LOG_LEVEL_PES) << "Emit a reassembled PES:"
+                       << " size=" << es_size << " pts=" << media_pts
+                       << " dts=" << media_dts << " data_alignment_indicator="
+                       << data_alignment_indicator;
   return es_parser_->Parse(&raw_pes[es_offset], es_size, media_pts, media_dts);
 }