X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_examiner.cc;h=32b60e0cba53e61009019c5bd217fe57c8766304;hb=2913a26dd799a6cda6f1b94415167362106318dc;hp=6c76ba5a433266591673cc34d645256fb6f1ed39;hpb=17f9254f68fabfe5d8136d256b69a65fc58724ee;p=dcpomatic.git

diff --git a/src/lib/ffmpeg_examiner.cc b/src/lib/ffmpeg_examiner.cc
index 6c76ba5a4..32b60e0cb 100644
--- a/src/lib/ffmpeg_examiner.cc
+++ b/src/lib/ffmpeg_examiner.cc
@@ -18,11 +18,14 @@
 
 */
 
+#include "dcpomatic_log.h"
 extern "C" {
 #include <libavcodec/avcodec.h>
 #include <libavformat/avformat.h>
 #include <libavutil/pixfmt.h>
 #include <libavutil/pixdesc.h>
+#include <libavutil/eval.h>
+#include <libavutil/display.h>
 }
 #include "ffmpeg_examiner.h"
 #include "ffmpeg_content.h"
@@ -38,20 +41,28 @@ extern "C" {
 using std::string;
 using std::cout;
 using std::max;
+using std::vector;
 using boost::shared_ptr;
 using boost::optional;
 
+/* 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];
-		if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
+		if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO && s->codec->codec) {
 
 			/* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
 			   so bodge it here.  No idea why we should have to do this.
@@ -62,7 +73,6 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
 			}
 
 			DCPOMATIC_ASSERT (_format_context->duration != AV_NOPTS_VALUE);
-			DCPOMATIC_ASSERT (s->codec->codec);
 			DCPOMATIC_ASSERT (s->codec->codec->name);
 
 			_audio_streams.push_back (
@@ -72,7 +82,7 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
 						s->codec->codec->name,
 						s->id,
 						s->codec->sample_rate,
-						(double (_format_context->duration) / AV_TIME_BASE) * s->codec->sample_rate,
+						llrint ((double (_format_context->duration) / AV_TIME_BASE) * s->codec->sample_rate),
 						s->codec->channels
 						)
 					)
@@ -98,9 +108,15 @@ 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) {
@@ -118,7 +134,7 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
 		AVCodecContext* context = _format_context->streams[_packet.stream_index]->codec;
 
 		if (_video_stream && _packet.stream_index == _video_stream.get()) {
-			video_packet (context);
+			video_packet (context, temporal_reference);
 		}
 
 		bool got_all_audio = true;
@@ -134,19 +150,60 @@ 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;
 		}
 	}
+
+	if (_video_stream) {
+		/* This code taken from get_rotation() in ffmpeg:cmdutils.c */
+		AVStream* stream = _format_context->streams[*_video_stream];
+		AVDictionaryEntry* rotate_tag = av_dict_get (stream->metadata, "rotate", 0, 0);
+		uint8_t* displaymatrix = av_stream_get_side_data (stream, AV_PKT_DATA_DISPLAYMATRIX, 0);
+		_rotation = 0;
+
+		if (rotate_tag && *rotate_tag->value && strcmp(rotate_tag->value, "0")) {
+			char *tail;
+			_rotation = av_strtod (rotate_tag->value, &tail);
+			if (*tail) {
+				_rotation = 0;
+			}
+		}
+
+		if (displaymatrix && !_rotation) {
+			_rotation = - av_display_rotation_get ((int32_t*) displaymatrix);
+		}
+
+		_rotation = *_rotation - 360 * floor (*_rotation / 360 + 0.9 / 360);
+	}
+
+	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) {
+		_id = e->value;
+	}
+#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;
 	}
 
@@ -160,6 +217,10 @@ 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");
+		}
 	}
 }