X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_examiner.cc;h=5683555e4aaffc42b3bce4493b17e9a5f1f3a72b;hb=5f1fdbafc6eef37523250e0b8542a8939a038823;hp=3fb9a53e4618f49e6f4aedb82780e9f10de541b3;hpb=92784c9c28c48859578cd6e75aa01d5657d0c341;p=dcpomatic.git

diff --git a/src/lib/ffmpeg_examiner.cc b/src/lib/ffmpeg_examiner.cc
index 3fb9a53e4..5683555e4 100644
--- a/src/lib/ffmpeg_examiner.cc
+++ b/src/lib/ffmpeg_examiner.cc
@@ -18,6 +18,15 @@
 
 */
 
+#include "dcpomatic_log.h"
+#include "ffmpeg_examiner.h"
+#include "ffmpeg_content.h"
+#include "job.h"
+#include "ffmpeg_audio_stream.h"
+#include "ffmpeg_subtitle_stream.h"
+#include "util.h"
+#include "warnings.h"
+DCPOMATIC_DISABLE_WARNINGS
 extern "C" {
 #include <libavcodec/avcodec.h>
 #include <libavformat/avformat.h>
@@ -26,12 +35,7 @@ extern "C" {
 #include <libavutil/eval.h>
 #include <libavutil/display.h>
 }
-#include "ffmpeg_examiner.h"
-#include "ffmpeg_content.h"
-#include "job.h"
-#include "ffmpeg_audio_stream.h"
-#include "ffmpeg_subtitle_stream.h"
-#include "util.h"
+DCPOMATIC_ENABLE_WARNINGS
 #include <boost/foreach.hpp>
 #include <iostream>
 
@@ -40,20 +44,30 @@ extern "C" {
 using std::string;
 using std::cout;
 using std::max;
+using std::vector;
 using boost::shared_ptr;
 using boost::optional;
 using namespace dcpomatic;
 
+
+/* This is how many frames from the start of any video that we will examine to see if we
+ * can spot soft 2:3 pull-down ("telecine").
+ */
+static const int PULLDOWN_CHECK_FRAMES = 16;
+
+
 /** @param job job that the examiner is operating in, or 0 */
 FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Job> job)
 	: FFmpeg (c)
 	, _video_length (0)
 	, _need_video_length (false)
+	, _pulldown (false)
 {
 	/* Find audio and subtitle streams */
 
 	for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
 		AVStream* s = _format_context->streams[i];
+DCPOMATIC_DISABLE_WARNINGS
 		if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
 
 			/* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
@@ -101,9 +115,15 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
 	/* Run through until we find:
 	 *   - the first video.
 	 *   - the first audio for each stream.
+	 *   - the top-field-first and repeat-first-frame values ("temporal_reference") for the first PULLDOWN_CHECK_FRAMES video frames.
 	 */
 
 	int64_t const len = _file_group.length ();
+	/* A string which we build up to describe the top-field-first and repeat-first-frame values for the first few frames.
+	 * It would be nicer to use something like vector<bool> here but we want to search the array for a pattern later,
+	 * and a string seems a reasonably neat way to do that.
+	 */
+	string temporal_reference;
 	while (true) {
 		int r = av_read_frame (_format_context, &_packet);
 		if (r < 0) {
@@ -119,9 +139,10 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
 		}
 
 		AVCodecContext* context = _format_context->streams[_packet.stream_index]->codec;
+DCPOMATIC_ENABLE_WARNINGS
 
 		if (_video_stream && _packet.stream_index == _video_stream.get()) {
-			video_packet (context);
+			video_packet (context, temporal_reference);
 		}
 
 		bool got_all_audio = true;
@@ -137,7 +158,7 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
 
 		av_packet_unref (&_packet);
 
-		if (_first_video && got_all_audio) {
+		if (_first_video && got_all_audio && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) {
 			/* All done */
 			break;
 		}
@@ -167,6 +188,13 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
 		DCPOMATIC_ASSERT (fabs (*_rotation - 90 * round (*_rotation / 90)) < 2);
 	}
 
+	LOG_GENERAL("Temporal reference was %1", temporal_reference);
+	if (temporal_reference.find("T2T3B2B3T2T3B2B3") != string::npos || temporal_reference.find("B2B3T2T3B2B3T2T3") != string::npos) {
+		/* The magical sequence (taken from mediainfo) suggests that 2:3 pull-down is in use */
+		_pulldown = true;
+		LOG_GENERAL_NC("Suggest that this may be 2:3 pull-down (soft telecine)");
+	}
+
 #ifdef DCPOMATIC_VARIANT_SWAROOP
 	AVDictionaryEntry* e = av_dict_get (_format_context->metadata, SWAROOP_ID_TAG, 0, 0);
 	if (e) {
@@ -175,17 +203,24 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
 #endif
 }
 
+
+/** @param temporal_reference A string to which we should add two characters per frame;
+ *  the first   is T or B depending on whether it's top- or bottom-field first,
+ *  ths seconds is 3 or 2 depending on whether "repeat_pict" is true or not.
+ */
 void
-FFmpegExaminer::video_packet (AVCodecContext* context)
+FFmpegExaminer::video_packet (AVCodecContext* context, string& temporal_reference)
 {
 	DCPOMATIC_ASSERT (_video_stream);
 
-	if (_first_video && !_need_video_length) {
+	if (_first_video && !_need_video_length && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) {
 		return;
 	}
 
 	int frame_finished;
+DCPOMATIC_DISABLE_WARNINGS
 	if (avcodec_decode_video2 (context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
+DCPOMATIC_ENABLE_WARNINGS
 		if (!_first_video) {
 			_first_video = frame_time (_format_context->streams[_video_stream.get()]);
 		}
@@ -194,9 +229,14 @@ FFmpegExaminer::video_packet (AVCodecContext* context)
 				_format_context->streams[_video_stream.get()]
 				).get_value_or (ContentTime ()).frames_round (video_frame_rate().get ());
 		}
+		if (temporal_reference.size() < (PULLDOWN_CHECK_FRAMES * 2)) {
+			temporal_reference += (_frame->top_field_first ? "T" : "B");
+			temporal_reference += (_frame->repeat_pict ? "3" : "2");
+		}
 	}
 }
 
+
 void
 FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr<FFmpegAudioStream> stream)
 {
@@ -205,17 +245,22 @@ FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr<FFmpegAudioStr
 	}
 
 	int frame_finished;
+DCPOMATIC_DISABLE_WARNINGS
 	if (avcodec_decode_audio4 (context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
+DCPOMATIC_ENABLE_WARNINGS
 		stream->first_audio = frame_time (stream->stream (_format_context));
 	}
 }
 
+
 optional<ContentTime>
 FFmpegExaminer::frame_time (AVStream* s) const
 {
 	optional<ContentTime> t;
 
+DCPOMATIC_DISABLE_WARNINGS
 	int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
+DCPOMATIC_ENABLE_WARNINGS
 	if (bet != AV_NOPTS_VALUE) {
 		t = ContentTime::from_seconds (bet * av_q2d (s->time_base));
 	}
@@ -223,6 +268,7 @@ FFmpegExaminer::frame_time (AVStream* s) const
 	return t;
 }
 
+
 optional<double>
 FFmpegExaminer::video_frame_rate () const
 {