X-Git-Url: https://git.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fvideo_decoder.cc;h=cf21f885a3fbc6af45ba2efde1cfc18ad645eb9b;hp=81991b88298f7c89c2571a656960abede3b1f9b7;hb=c103d8c1306e5fb3937b3a6c430a3fff32653fa3;hpb=3828baf56467224f5d44049bf1e7a7ed11f43a05 diff --git a/src/lib/video_decoder.cc b/src/lib/video_decoder.cc index 81991b882..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. @@ -18,346 +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) -#ifdef DCPOMATIC_DEBUG - : test_gaps (0) - , _parent (parent), - _content (c) -#else - : _parent (parent) - , _content (c) -#endif - , _log (log) - , _last_seek_accurate (true) - , _ignore (false) -{ - _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 == 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. - */ -list -VideoDecoder::get (Frame frame, bool accurate) -{ - if (_no_data_frame && frame >= _no_data_frame.get()) { - return list (); - } - - /* At this stage, if we have get_video()ed before, _decoded will contain the last frame that this - method returned (and possibly a few more). If the requested frame is not in _decoded and it is not the next - one after the end of _decoded we need to seek. - */ - - _log->log (String::compose ("VD has request for %1", frame), LogEntry::TYPE_DEBUG_DECODE); - - if (_decoded.empty() || frame < _decoded.front().frame || frame > (_decoded.back().frame + 1)) { - _parent->seek (ContentTime::from_frames (frame, _content->active_video_frame_rate()), accurate); - } - 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).empty ()) { - /* 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; - } - - if (!_decoded.empty() && _decoded.front().frame > 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). - */ - break; - } - } - - dec = decoded (frame); - - if (no_data && dec.empty()) { - _no_data_frame = frame; - } - - } else { - /* Any frame will do: use the first one that comes out of pass() */ - while (_decoded.empty() && !_parent->pass (Decoder::PASS_REASON_VIDEO, accurate)) {} - if (!_decoded.empty ()) { - dec.push_back (_decoded.front ()); - } - } - - /* 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 < dec.front().frame) { - _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) +/** 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::fill_one_eye (Frame from, Frame to, Eyes eye) +VideoDecoder::emit (shared_ptr film, shared_ptr image, Frame decoder_frame) { - if (to == 0) { - /* Already OK */ + if (ignore ()) { 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, eye, filler_part, i) - ); - } -} - -/** Fill _decoded from `from' up to, but not including, `to' - * adding both left and right eye frames. - */ -void -VideoDecoder::fill_both_eyes (Frame from, Frame to, Eyes eye) -{ - if (to == 0 && eye == EYES_LEFT) { - /* Already OK */ - return; - } + auto const afr = _content->active_video_frame_rate(film); + auto const vft = _content->video->frame_type(); - /* 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; + auto frame_time = ContentTime::from_frames (decoder_frame, afr); - /* ...unless there's some video we can fill with */ - for (list::const_reverse_iterator i = _decoded.rbegin(); i != _decoded.rend(); ++i) { - if (i->eyes == EYES_LEFT && !filler_left_image) { - filler_left_image = i->image; - filler_left_part = i->part; - } else if (i->eyes == EYES_RIGHT && !filler_right_image) { - filler_right_image = i->image; - filler_right_part = i->part; + /* 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) + ) + ) + ); } - if (filler_left_image && filler_right_image) { - break; + if (_frame_interval_checker->guess() != FrameIntervalChecker::AGAIN) { + _frame_interval_checker.reset (); } } - Frame filler_frame = from; - Eyes filler_eye = _decoded.empty() ? EYES_LEFT : _decoded.back().eyes; - - if (_decoded.empty ()) { - filler_frame = 0; - filler_eye = EYES_LEFT; - } else if (_decoded.back().eyes == EYES_LEFT) { - filler_frame = _decoded.back().frame; - filler_eye = EYES_RIGHT; - } else if (_decoded.back().eyes == EYES_RIGHT) { - filler_frame = _decoded.back().frame + 1; - filler_eye = EYES_LEFT; - } - - while (filler_frame != to || filler_eye != eye) { - -#ifdef DCPOMATIC_DEBUG - test_gaps++; -#endif - - _decoded.push_back ( - ContentVideo ( - filler_eye == EYES_LEFT ? filler_left_image : filler_right_image, - filler_eye, - filler_eye == EYES_LEFT ? filler_left_part : filler_right_part, - filler_frame - ) - ); + 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 (filler_eye == EYES_LEFT) { - filler_eye = EYES_RIGHT; + if (vft == VideoFrameType::THREE_D_ALTERNATE) { + frame = decoder_frame / 2; + eyes = (decoder_frame % 2) ? Eyes::RIGHT : Eyes::LEFT; } else { - filler_eye = EYES_LEFT; - ++filler_frame; + 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. + */ + if (j2k && j2k->eye()) { + eyes = j2k->eye().get() == dcp::Eye::LEFT ? Eyes::LEFT : Eyes::RIGHT; + } + } } - } -} -/** Called by decoder classes when they have a video frame ready */ -void -VideoDecoder::give (shared_ptr image, Frame frame) -{ - if (_ignore) { - return; + _position = ContentTime::from_frames (frame, afr); + } else { + 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; + } } - _log->log (String::compose ("VD receives %1", frame), LogEntry::TYPE_DEBUG_DECODE); - - /* 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, EYES_BOTH, PART_WHOLE, frame)); + switch (vft) { + case VideoFrameType::TWO_D: + case VideoFrameType::THREE_D: + Data (ContentVideo (image, frame, eyes, Part::WHOLE)); break; - case VIDEO_FRAME_TYPE_3D_ALTERNATE: + case VideoFrameType::THREE_D_ALTERNATE: { - /* We receive the same frame index twice for 3D-alternate; hence we know which - frame this one is. - */ - bool const same = (!_decoded.empty() && frame == _decoded.back().frame); - to_push.push_back (ContentVideo (image, same ? EYES_RIGHT : EYES_LEFT, PART_WHOLE, frame)); + Data (ContentVideo (image, frame, eyes, Part::WHOLE)); + _last_emitted_eyes = eyes; break; } - case VIDEO_FRAME_TYPE_3D_LEFT_RIGHT: - to_push.push_back (ContentVideo (image, EYES_LEFT, PART_LEFT_HALF, frame)); - to_push.push_back (ContentVideo (image, EYES_RIGHT, PART_RIGHT_HALF, frame)); + 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_TOP_BOTTOM: - to_push.push_back (ContentVideo (image, EYES_LEFT, PART_TOP_HALF, frame)); - to_push.push_back (ContentVideo (image, EYES_RIGHT, PART_BOTTOM_HALF, frame)); + 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_LEFT: - to_push.push_back (ContentVideo (image, EYES_LEFT, PART_WHOLE, frame)); + case VideoFrameType::THREE_D_LEFT: + Data (ContentVideo (image, frame, Eyes::LEFT, Part::WHOLE)); break; - case VIDEO_FRAME_TYPE_3D_RIGHT: - to_push.push_back (ContentVideo (image, EYES_RIGHT, PART_WHOLE, frame)); + 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; - optional to; - - if (_decoded.empty() && _last_seek_time && _last_seek_accurate) { - from = _last_seek_time->frames_round (_content->active_video_frame_rate ()); - to = to_push.front().frame; - } else if (!_decoded.empty ()) { - from = _decoded.back().frame + 1; - to = to_push.front().frame; - } - - /* 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 && to && from.get() > to.get()) { - return; - } - - if (from) { - switch (_content->video->frame_type ()) { - case VIDEO_FRAME_TYPE_2D: - fill_one_eye (from.get(), to.get (), EYES_BOTH); - break; - 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.get(), to_push.front().eyes); - break; - case VIDEO_FRAME_TYPE_3D_LEFT: - fill_one_eye (from.get(), to.get (), EYES_LEFT); - break; - case VIDEO_FRAME_TYPE_3D_RIGHT: - fill_one_eye (from.get(), to.get (), 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() > 96) { - _decoded.pop_back (); - } + _position = ContentTime::from_frames (frame, afr); } -void -VideoDecoder::seek (ContentTime s, bool accurate) -{ - _decoded.clear (); - _last_seek_time = s; - _last_seek_accurate = accurate; -} -/** Set this decoder never to produce any data */ void -VideoDecoder::set_ignore () +VideoDecoder::seek () { - _ignore = true; + _position = boost::none; + _last_emitted_eyes.reset (); + _frame_interval_checker.reset (new FrameIntervalChecker()); }