Change how video timing is done.
[dcpomatic.git] / src / lib / video_decoder.cc
index 508ed90b71c71f62e874f0a763b4b19f2d9db8d7..c628fddd983e2015a81b1bd7d13b52b0d9424a0a 100644 (file)
@@ -1,5 +1,5 @@
 /*
-    Copyright (C) 2012-2018 Carl Hetherington <cth@carlh.net>
+    Copyright (C) 2012-2021 Carl Hetherington <cth@carlh.net>
 
     This file is part of DCP-o-matic.
 
 
 */
 
-#include "video_decoder.h"
+
+#include "compose.hpp"
+#include "frame_interval_checker.h"
 #include "image.h"
-#include "raw_image_proxy.h"
-#include "film.h"
+#include "j2k_image_proxy.h"
 #include "log.h"
-#include "compose.hpp"
-#include <boost/foreach.hpp>
+#include "raw_image_proxy.h"
+#include "video_decoder.h"
 #include <iostream>
 
 #include "i18n.h"
 
+
 using std::cout;
-using std::list;
-using std::max;
-using std::back_inserter;
-using boost::shared_ptr;
-using boost::optional;
+using std::dynamic_pointer_cast;
+using std::shared_ptr;
 using namespace dcpomatic;
 
+
 VideoDecoder::VideoDecoder (Decoder* parent, shared_ptr<const Content> c)
        : DecoderPart (parent)
        , _content (c)
+       , _frame_interval_checker (new FrameIntervalChecker())
 {
 
 }
 
-/** Called by decoder classes when they have a video frame ready.
- *  @param frame Frame index within the content; this does not take into account 3D
- *  so for 3D_ALTERNATE this value goes:
- *     0: frame 0 left
- *     1: frame 0 right
- *     2: frame 1 left
- *     3: frame 1 right
- *  and so on.
- */
+
+/** Called by decoder classes when they have a video frame ready */
 void
-VideoDecoder::emit (shared_ptr<const Film> film, shared_ptr<const ImageProxy> image, Frame decoder_frame)
+VideoDecoder::emit(shared_ptr<const Film> film, shared_ptr<const ImageProxy> image, ContentTime time)
 {
        if (ignore ()) {
                return;
        }
 
-       /* Before we `re-write' the frame indexes of these incoming data we need to check for
-          the case where the user has some 2D content which they have marked as 3D.  With 3D
-          we should get two frames for each frame index, but in this `bad' case we only get
-          one.  We need to throw an exception if this happens.
-       */
-
-       if (_content->video->frame_type() == VIDEO_FRAME_TYPE_3D) {
-               if (_last_threed_frames.size() > 4) {
-                       _last_threed_frames.erase (_last_threed_frames.begin());
-               }
-               _last_threed_frames.push_back (decoder_frame);
-               if (_last_threed_frames.size() == 4) {
-                       if (_last_threed_frames[0] != _last_threed_frames[1] || _last_threed_frames[2] != _last_threed_frames[3]) {
-                               boost::throw_exception (
-                                       DecodeError(
-                                               String::compose(
-                                                       _("The content file %1 is set as 3D but does not appear to contain 3D images.  Please set it to 2D.  "
-                                                         "You can still make a 3D DCP from this content by ticking the 3D option in the DCP video tab."),
-                                                       _content->path(0)
-                                                       )
+       auto const afr = _content->active_video_frame_rate(film);
+       auto const vft = _content->video->frame_type();
+
+       /* Do some heuristics to try and spot the case where the user sets content to 3D
+        * when it is not.  We try to tell this by looking at the differences in time between
+        * the first few frames.  Real 3D content should have two frames for each timestamp.
+        */
+       if (_frame_interval_checker) {
+               _frame_interval_checker->feed(time, afr);
+               if (_frame_interval_checker->guess() == FrameIntervalChecker::PROBABLY_NOT_3D && vft == VideoFrameType::THREE_D) {
+                       boost::throw_exception (
+                               DecodeError(
+                                       String::compose(
+                                               _("The content file %1 is set as 3D but does not appear to contain 3D images.  Please set it to 2D.  "
+                                                 "You can still make a 3D DCP from this content by ticking the 3D option in the DCP video tab."),
+                                               _content->path(0)
                                                )
-                                       );
-                       }
+                                       )
+                               );
                }
-       }
 
-       double const afr = _content->active_video_frame_rate(film);
-
-       Frame frame;
-       Eyes eyes = EYES_BOTH;
-       if (!_position) {
-               /* This is the first data we have received since initialisation or seek.  Set
-                  the position based on the frame that was given.  After this first time
-                  we just cound frames, since (as with audio) it seems that ContentTimes
-                  are unreliable from FFmpegDecoder.  They are much better than audio times
-                  but still we get the occasional one which is duplicated.  In this case
-                  ffmpeg seems to carry on regardless, processing the video frame as normal.
-                  If we drop the frame with the duplicated timestamp we obviously lose sync.
-               */
-               _position = ContentTime::from_frames (decoder_frame, afr);
-               if (_content->video->frame_type() == VIDEO_FRAME_TYPE_3D_ALTERNATE) {
-                       frame = decoder_frame / 2;
-                       _last_emitted_eyes = EYES_RIGHT;
-               } else {
-                       frame = decoder_frame;
-               }
-       } else {
-               VideoFrameType const ft = _content->video->frame_type ();
-               if (ft == VIDEO_FRAME_TYPE_3D_ALTERNATE || ft == VIDEO_FRAME_TYPE_3D) {
-                       DCPOMATIC_ASSERT (_last_emitted_eyes);
-                       if (_last_emitted_eyes.get() == EYES_RIGHT) {
-                               frame = _position->frames_round(afr) + 1;
-                               eyes = EYES_LEFT;
-                       } else {
-                               frame = _position->frames_round(afr);
-                               eyes = EYES_RIGHT;
-                       }
-               } else {
-                       frame = _position->frames_round(afr) + 1;
+               if (_frame_interval_checker->guess() != FrameIntervalChecker::AGAIN) {
+                       _frame_interval_checker.reset ();
                }
        }
 
-       switch (_content->video->frame_type ()) {
-       case VIDEO_FRAME_TYPE_2D:
-               Data (ContentVideo (image, frame, EYES_BOTH, PART_WHOLE));
+       switch (vft) {
+       case VideoFrameType::TWO_D:
+               Data(ContentVideo(image, time, Eyes::BOTH, Part::WHOLE));
                break;
-       case VIDEO_FRAME_TYPE_3D:
+       case VideoFrameType::THREE_D:
        {
-               Data (ContentVideo (image, frame, eyes, PART_WHOLE));
-               _last_emitted_frame = frame;
-               _last_emitted_eyes = eyes;
+               auto eyes = Eyes::LEFT;
+               auto j2k = dynamic_pointer_cast<const J2KImageProxy>(image);
+               if (j2k && j2k->eye()) {
+                       eyes = *j2k->eye() == dcp::Eye::LEFT ? Eyes::LEFT : Eyes::RIGHT;
+               }
+
+               Data(ContentVideo(image, time, eyes, Part::WHOLE));
                break;
        }
-       case VIDEO_FRAME_TYPE_3D_ALTERNATE:
+       case VideoFrameType::THREE_D_ALTERNATE:
        {
-               Data (ContentVideo (image, frame, eyes, PART_WHOLE));
+               Eyes eyes;
+               if (_last_emitted_eyes) {
+                       eyes = _last_emitted_eyes.get() == Eyes::LEFT ? Eyes::RIGHT : Eyes::LEFT;
+               } else {
+                       /* We don't know what eye this frame is, so just guess */
+                       auto frame = time.frames_round(_content->video_frame_rate().get_value_or(24));
+                       eyes = (frame % 2) ? Eyes::RIGHT : Eyes::LEFT;
+               }
+               Data(ContentVideo(image, time, eyes, Part::WHOLE));
                _last_emitted_eyes = eyes;
                break;
        }
-       case VIDEO_FRAME_TYPE_3D_LEFT_RIGHT:
-               Data (ContentVideo (image, frame, EYES_LEFT, PART_LEFT_HALF));
-               Data (ContentVideo (image, frame, EYES_RIGHT, PART_RIGHT_HALF));
+       case VideoFrameType::THREE_D_LEFT_RIGHT:
+               Data(ContentVideo(image, time, Eyes::LEFT, Part::LEFT_HALF));
+               Data(ContentVideo(image, time, Eyes::RIGHT, Part::RIGHT_HALF));
                break;
-       case VIDEO_FRAME_TYPE_3D_TOP_BOTTOM:
-               Data (ContentVideo (image, frame, EYES_LEFT, PART_TOP_HALF));
-               Data (ContentVideo (image, frame, EYES_RIGHT, PART_BOTTOM_HALF));
+       case VideoFrameType::THREE_D_TOP_BOTTOM:
+               Data(ContentVideo(image, time, Eyes::LEFT, Part::TOP_HALF));
+               Data(ContentVideo(image, time, Eyes::RIGHT, Part::BOTTOM_HALF));
                break;
-       case VIDEO_FRAME_TYPE_3D_LEFT:
-               Data (ContentVideo (image, frame, EYES_LEFT, PART_WHOLE));
+       case VideoFrameType::THREE_D_LEFT:
+               Data(ContentVideo(image, time, Eyes::LEFT, Part::WHOLE));
                break;
-       case VIDEO_FRAME_TYPE_3D_RIGHT:
-               Data (ContentVideo (image, frame, EYES_RIGHT, PART_WHOLE));
+       case VideoFrameType::THREE_D_RIGHT:
+               Data(ContentVideo(image, time, Eyes::RIGHT, Part::WHOLE));
                break;
        default:
                DCPOMATIC_ASSERT (false);
        }
 
-       _position = ContentTime::from_frames (frame, afr);
+       _position = time;
 }
 
+
 void
 VideoDecoder::seek ()
 {
-       _position = boost::optional<ContentTime>();
-       _last_emitted_frame.reset ();
+       _position = boost::none;
        _last_emitted_eyes.reset ();
+       _frame_interval_checker.reset (new FrameIntervalChecker());
 }