X-Git-Url: https://git.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fvideo_decoder.cc;h=cf21f885a3fbc6af45ba2efde1cfc18ad645eb9b;hp=0e9ee0c1a874d1e9ffe1f7e0531b15369f431d83;hb=c103d8c1306e5fb3937b3a6c430a3fff32653fa3;hpb=936ff6927ad24daf0ed20776b19b6faa3df9bf83 diff --git a/src/lib/video_decoder.cc b/src/lib/video_decoder.cc index 0e9ee0c1a..cf21f885a 100644 --- a/src/lib/video_decoder.cc +++ b/src/lib/video_decoder.cc @@ -1,5 +1,5 @@ /* - Copyright (C) 2012-2016 Carl Hetherington + Copyright (C) 2012-2021 Carl Hetherington This file is part of DCP-o-matic. @@ -18,377 +18,174 @@ */ -#include "video_decoder.h" -#include "image.h" -#include "raw_image_proxy.h" + +#include "compose.hpp" #include "film.h" +#include "frame_interval_checker.h" +#include "image.h" +#include "j2k_image_proxy.h" #include "log.h" -#include "compose.hpp" -#include +#include "raw_image_proxy.h" +#include "video_decoder.h" #include #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 c, shared_ptr log) - : DecoderPart (parent, log) -#ifdef DCPOMATIC_DEBUG - , test_gaps (0) -#endif - , _content (c) - , _last_seek_accurate (true) -{ - _black_image.reset (new Image (AV_PIX_FMT_RGB24, _content->video->size(), true)); - _black_image->make_black (); -} -list -VideoDecoder::decoded (Frame frame) +VideoDecoder::VideoDecoder (Decoder* parent, shared_ptr c) + : DecoderPart (parent) + , _content (c) + , _frame_interval_checker (new FrameIntervalChecker()) { - list output; - BOOST_FOREACH (ContentVideo const & i, _decoded) { - if (i.frame.index() == frame) { - output.push_back (i); - } - } - - return output; } -/** Get all frames which exist in the content at a given frame index. - * @param frame Frame index. - * @param accurate true to try hard to return frames at the precise time that was requested, otherwise frames nearby may be returned. - * @return Frames; there may be none (if there is no video there), 1 for 2D or 2 for 3D. + +/** 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. */ -list -VideoDecoder::get (Frame frame, bool accurate) +void +VideoDecoder::emit (shared_ptr film, shared_ptr image, Frame decoder_frame) { - if (_no_data_frame && frame >= _no_data_frame.get()) { - return list (); + if (ignore ()) { + return; } - _log->log (String::compose ("VD has request for %1", frame), LogEntry::TYPE_DEBUG_DECODE); + auto const afr = _content->active_video_frame_rate(film); + auto const vft = _content->video->frame_type(); - /* See if we have frame, and suggest a seek if not */ + auto frame_time = ContentTime::from_frames (decoder_frame, afr); - list::const_iterator i = _decoded.begin (); - while (i != _decoded.end() && i->frame.index() != frame) { - _log->log (String::compose ("VD has stored %1 which is no good", i->frame.index()), LogEntry::TYPE_DEBUG_DECODE); - ++i; - } - - if (i == _decoded.end()) { - Frame seek_frame = frame; - if (_content->video->frame_type() == VIDEO_FRAME_TYPE_3D_ALTERNATE) { - /* 3D alternate is a special case as the frame index in the content is not the same - as the frame index we are talking about here. - */ - seek_frame *= 2; + /* 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 (frame_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) + ) + ) + ); } - _log->log (String::compose ("VD suggests seek to %1", seek_frame), LogEntry::TYPE_DEBUG_DECODE); - maybe_seek (ContentTime::from_frames (seek_frame, _content->active_video_frame_rate()), accurate); - } - /* Work out the number of frames that we should return; we - must return all frames in our content at the requested `time' - (i.e. frame) - */ - unsigned int frames_wanted = 0; - switch (_content->video->frame_type()) { - case VIDEO_FRAME_TYPE_2D: - case VIDEO_FRAME_TYPE_3D_LEFT: - case VIDEO_FRAME_TYPE_3D_RIGHT: - frames_wanted = 1; - break; - case VIDEO_FRAME_TYPE_3D: - case VIDEO_FRAME_TYPE_3D_ALTERNATE: - case VIDEO_FRAME_TYPE_3D_LEFT_RIGHT: - case VIDEO_FRAME_TYPE_3D_TOP_BOTTOM: - frames_wanted = 2; - break; - default: - DCPOMATIC_ASSERT (false); + if (_frame_interval_checker->guess() != FrameIntervalChecker::AGAIN) { + _frame_interval_checker.reset (); + } } - list dec; - - /* Now enough pass() calls should either: - * (a) give us what we want, or - * (b) give us something after what we want, indicating that we will never get what we want, or - * (c) hit the end of the decoder. - */ - if (accurate) { - /* We are being accurate, so we want the right frame. - * This could all be one statement but it's split up for clarity. - */ - bool no_data = false; - - while (true) { - if (decoded(frame).size() == frames_wanted) { - /* We got what we want */ - break; - } - - if (_parent->pass (Decoder::PASS_REASON_VIDEO, accurate)) { - /* The decoder has nothing more for us */ - no_data = true; - break; - } + 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 count 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. + */ - if (!_decoded.empty() && _decoded.front().frame.index() > frame) { - /* We're never going to get the frame we want. Perhaps the caller is asking - * for a video frame before the content's video starts (if its audio - * begins before its video, for example). + if (vft == VideoFrameType::THREE_D_ALTERNATE) { + frame = decoder_frame / 2; + eyes = (decoder_frame % 2) ? Eyes::RIGHT : Eyes::LEFT; + } else { + frame = decoder_frame; + if (vft == VideoFrameType::THREE_D) { + auto j2k = dynamic_pointer_cast(image); + /* At the moment only DCP decoders producers VideoFrameType::THREE_D, so only the J2KImageProxy + * knows which eye it is. */ - break; + if (j2k && j2k->eye()) { + eyes = j2k->eye().get() == dcp::Eye::LEFT ? Eyes::LEFT : Eyes::RIGHT; + } } } - dec = decoded (frame); - - if (no_data && dec.empty()) { - _no_data_frame = frame; - } - + _position = ContentTime::from_frames (frame, afr); } else { - /* Any frame(s) will do: use the first one(s) that comes out of pass() */ - while (_decoded.size() < frames_wanted && !_parent->pass (Decoder::PASS_REASON_VIDEO, accurate)) {} - list::const_iterator i = _decoded.begin(); - unsigned int j = 0; - while (i != _decoded.end() && j < frames_wanted) { - dec.push_back (*i); - ++i; - ++j; - } - } - - /* Clean up _decoded; keep the frame we are returning, if any (which may have two images - for 3D), but nothing before that - */ - while (!_decoded.empty() && !dec.empty() && _decoded.front().frame.index() < dec.front().frame.index()) { - _log->log (String::compose ("VD discards %1", _decoded.front().frame.index()), LogEntry::TYPE_DEBUG_DECODE); - _decoded.pop_front (); - } - - return dec; -} - -/** Fill _decoded from `from' up to, but not including, `to' with - * a frame for one particular Eyes value (which could be EYES_BOTH, - * EYES_LEFT or EYES_RIGHT) - */ -void -VideoDecoder::fill_one_eye (Frame from, Frame to, Eyes eye) -{ - if (to == 0) { - /* Already OK */ - return; - } - - /* Fill with black... */ - shared_ptr filler_image (new RawImageProxy (_black_image)); - Part filler_part = PART_WHOLE; - - /* ...unless there's some video we can fill with */ - if (!_decoded.empty ()) { - filler_image = _decoded.back().image; - filler_part = _decoded.back().part; - } - - for (Frame i = from; i < to; ++i) { -#ifdef DCPOMATIC_DEBUG - test_gaps++; -#endif - _decoded.push_back ( - ContentVideo (filler_image, VideoFrame (i, eye), filler_part) - ); - } -} - -/** Fill _decoded from `from' up to, but not including, `to' - * adding both left and right eye frames. - */ -void -VideoDecoder::fill_both_eyes (VideoFrame from, VideoFrame to) -{ - /* Fill with black... */ - shared_ptr filler_left_image (new RawImageProxy (_black_image)); - shared_ptr filler_right_image (new RawImageProxy (_black_image)); - Part filler_left_part = PART_WHOLE; - Part filler_right_part = PART_WHOLE; - - /* ...unless there's some video we can fill with */ - for (list::const_reverse_iterator i = _decoded.rbegin(); i != _decoded.rend(); ++i) { - if (i->frame.eyes() == EYES_LEFT && !filler_left_image) { - filler_left_image = i->image; - filler_left_part = i->part; - } else if (i->frame.eyes() == EYES_RIGHT && !filler_right_image) { - filler_right_image = i->image; - filler_right_part = i->part; - } - - if (filler_left_image && filler_right_image) { - break; + if (vft == VideoFrameType::THREE_D) { + auto j2k = dynamic_pointer_cast(image); + if (j2k && j2k->eye()) { + if (j2k->eye() == dcp::Eye::LEFT) { + frame = _position->frames_round(afr) + 1; + eyes = Eyes::LEFT; + } else { + frame = _position->frames_round(afr); + eyes = Eyes::RIGHT; + } + } else { + /* This should not happen; see above */ + frame = _position->frames_round(afr) + 1; + } + } else if (vft == VideoFrameType::THREE_D_ALTERNATE) { + 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; } } - while (from != to) { - -#ifdef DCPOMATIC_DEBUG - test_gaps++; -#endif - - _decoded.push_back ( - ContentVideo ( - from.eyes() == EYES_LEFT ? filler_left_image : filler_right_image, - from, - from.eyes() == EYES_LEFT ? filler_left_part : filler_right_part - ) - ); - - ++from; - } -} - -/** 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. - */ -void -VideoDecoder::give (shared_ptr image, Frame frame) -{ - if (ignore ()) { - return; - } - - _log->log (String::compose ("VD receives %1", frame), LogEntry::TYPE_DEBUG_DECODE); - _position = ContentTime::from_frames (frame, _content->active_video_frame_rate()); - - /* Work out what we are going to push into _decoded next */ - list to_push; - switch (_content->video->frame_type ()) { - case VIDEO_FRAME_TYPE_2D: - to_push.push_back (ContentVideo (image, VideoFrame (frame, EYES_BOTH), PART_WHOLE)); + switch (vft) { + case VideoFrameType::TWO_D: + case VideoFrameType::THREE_D: + Data (ContentVideo (image, frame, eyes, Part::WHOLE)); break; - case VIDEO_FRAME_TYPE_3D: + case VideoFrameType::THREE_D_ALTERNATE: { - /* We receive the same frame index twice for 3D; hence we know which - frame this one is. - */ - bool const same = (!_decoded.empty() && frame == _decoded.back().frame.index()); - to_push.push_back (ContentVideo (image, VideoFrame (frame, same ? EYES_RIGHT : EYES_LEFT), PART_WHOLE)); + Data (ContentVideo (image, frame, eyes, Part::WHOLE)); + _last_emitted_eyes = eyes; break; } - case VIDEO_FRAME_TYPE_3D_ALTERNATE: - to_push.push_back (ContentVideo (image, VideoFrame (frame / 2, (frame % 2) ? EYES_RIGHT : EYES_LEFT), PART_WHOLE)); + case VideoFrameType::THREE_D_LEFT_RIGHT: + Data (ContentVideo (image, frame, Eyes::LEFT, Part::LEFT_HALF)); + Data (ContentVideo (image, frame, Eyes::RIGHT, Part::RIGHT_HALF)); break; - case VIDEO_FRAME_TYPE_3D_LEFT_RIGHT: - to_push.push_back (ContentVideo (image, VideoFrame (frame, EYES_LEFT), PART_LEFT_HALF)); - to_push.push_back (ContentVideo (image, VideoFrame (frame, EYES_RIGHT), PART_RIGHT_HALF)); + case VideoFrameType::THREE_D_TOP_BOTTOM: + Data (ContentVideo (image, frame, Eyes::LEFT, Part::TOP_HALF)); + Data (ContentVideo (image, frame, Eyes::RIGHT, Part::BOTTOM_HALF)); break; - case VIDEO_FRAME_TYPE_3D_TOP_BOTTOM: - to_push.push_back (ContentVideo (image, VideoFrame (frame, EYES_LEFT), PART_TOP_HALF)); - to_push.push_back (ContentVideo (image, VideoFrame (frame, EYES_RIGHT), PART_BOTTOM_HALF)); + case VideoFrameType::THREE_D_LEFT: + Data (ContentVideo (image, frame, Eyes::LEFT, Part::WHOLE)); break; - case VIDEO_FRAME_TYPE_3D_LEFT: - to_push.push_back (ContentVideo (image, VideoFrame (frame, EYES_LEFT), PART_WHOLE)); - break; - case VIDEO_FRAME_TYPE_3D_RIGHT: - to_push.push_back (ContentVideo (image, VideoFrame (frame, EYES_RIGHT), PART_WHOLE)); + case VideoFrameType::THREE_D_RIGHT: + Data (ContentVideo (image, frame, Eyes::RIGHT, Part::WHOLE)); break; default: DCPOMATIC_ASSERT (false); } - /* Now VideoDecoder is required never to have gaps in the frames that it presents - via get_video(). Hence we need to fill in any gap between the last thing in _decoded - and the things we are about to push. - */ - - optional from; - - if (_decoded.empty() && _last_seek_time && _last_seek_accurate) { - from = VideoFrame ( - _last_seek_time->frames_round (_content->active_video_frame_rate ()), - _content->video->frame_type() == VIDEO_FRAME_TYPE_2D ? EYES_BOTH : EYES_LEFT - ); - } else if (!_decoded.empty ()) { - /* Get the last frame we have */ - from = _decoded.back().frame; - /* And move onto the first frame we need */ - ++(*from); - if (_content->video->frame_type() == VIDEO_FRAME_TYPE_3D_LEFT || _content->video->frame_type() == VIDEO_FRAME_TYPE_3D_RIGHT) { - /* The previous ++ will increment a 3D-left-eye to the same index right-eye. If we are dealing with - a single-eye source we need an extra ++ to move back to the same eye. - */ - ++(*from); - } - } - - /* If we've pre-rolled on a seek we may now receive out-of-order frames - (frames before the last seek time) which we can just ignore. - */ - if (from && (*from) > to_push.front().frame) { - return; - } - - unsigned int const max_decoded_size = 96; - - /* If _decoded is already `full' there is no point in adding anything more to it, - as the new stuff will just be removed again. - */ - if (_decoded.size() < max_decoded_size) { - if (from) { - switch (_content->video->frame_type ()) { - case VIDEO_FRAME_TYPE_2D: - fill_one_eye (from->index(), to_push.front().frame.index(), EYES_BOTH); - break; - case VIDEO_FRAME_TYPE_3D: - case VIDEO_FRAME_TYPE_3D_LEFT_RIGHT: - case VIDEO_FRAME_TYPE_3D_TOP_BOTTOM: - case VIDEO_FRAME_TYPE_3D_ALTERNATE: - fill_both_eyes (from.get(), to_push.front().frame); - break; - case VIDEO_FRAME_TYPE_3D_LEFT: - fill_one_eye (from->index(), to_push.front().frame.index(), EYES_LEFT); - break; - case VIDEO_FRAME_TYPE_3D_RIGHT: - fill_one_eye (from->index(), to_push.front().frame.index(), EYES_RIGHT); - break; - } - } - - copy (to_push.begin(), to_push.end(), back_inserter (_decoded)); - } - - /* We can't let this build up too much or we will run out of memory. There is a - `best' value for the allowed size of _decoded which balances memory use - with decoding efficiency (lack of seeks). Throwing away video frames here - is not a problem for correctness, so do it. - */ - while (_decoded.size() > max_decoded_size) { - _decoded.pop_back (); - } + _position = ContentTime::from_frames (frame, afr); } + void -VideoDecoder::seek (ContentTime s, bool accurate) +VideoDecoder::seek () { - _log->log (String::compose ("VD seek to %1", to_string(s)), LogEntry::TYPE_DEBUG_DECODE); - _decoded.clear (); - _last_seek_time = s; - _last_seek_accurate = accurate; - _position.reset (); + _position = boost::none; + _last_emitted_eyes.reset (); + _frame_interval_checker.reset (new FrameIntervalChecker()); }