/*
- Copyright (C) 2012-2015 Carl Hetherington <cth@carlh.net>
+ Copyright (C) 2012-2021 Carl Hetherington <cth@carlh.net>
- This program is free software; you can redistribute it and/or modify
+ This file is part of DCP-o-matic.
+
+ DCP-o-matic is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
- This program is distributed in the hope that it will be useful,
+ DCP-o-matic is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
- along with this program; if not, write to the Free Software
- Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ along with DCP-o-matic. If not, see <http://www.gnu.org/licenses/>.
*/
-#include "video_decoder.h"
-#include "image.h"
-#include "image_proxy.h"
-#include "raw_image_proxy.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 "raw_image_proxy.h"
+#include "video_decoder.h"
+#include <iostream>
#include "i18n.h"
+
+using std::back_inserter;
using std::cout;
+using std::dynamic_pointer_cast;
using std::list;
+using std::make_shared;
using std::max;
-using std::back_inserter;
-using boost::shared_ptr;
+using std::shared_ptr;
using boost::optional;
+using namespace dcpomatic;
-VideoDecoder::VideoDecoder (shared_ptr<const VideoContent> c)
-#ifdef DCPOMATIC_DEBUG
- : test_gaps (0)
- , _video_content (c)
-#else
- : _video_content (c)
-#endif
- , _last_seek_accurate (true)
- , _ignore_video (false)
-{
- _black_image.reset (new Image (PIX_FMT_RGB24, _video_content->video_size(), true));
- _black_image->make_black ();
-}
-list<ContentVideo>
-VideoDecoder::decoded_video (Frame frame)
+VideoDecoder::VideoDecoder (Decoder* parent, shared_ptr<const Content> c)
+ : DecoderPart (parent)
+ , _content (c)
+ , _frame_interval_checker (new FrameIntervalChecker())
{
- list<ContentVideo> output;
-
- for (list<ContentVideo>::const_iterator i = _decoded_video.begin(); i != _decoded_video.end(); ++i) {
- 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<ContentVideo>
-VideoDecoder::get_video (Frame frame, bool accurate)
-{
- /* At this stage, if we have get_video()ed before, _decoded_video will contain the last frame that this
- method returned (and possibly a few more). If the requested frame is not in _decoded_video and it is not the next
- one after the end of _decoded_video we need to seek.
- */
-
- _video_content->film()->log()->log (String::compose ("VD has request for %1", frame), Log::TYPE_DEBUG_DECODE);
-
- if (_decoded_video.empty() || frame < _decoded_video.front().frame || frame > (_decoded_video.back().frame + 1)) {
- seek (ContentTime::from_frames (frame, _video_content->video_frame_rate()), accurate);
- }
-
- list<ContentVideo> 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.
- */
- while (true) {
- if (!decoded_video(frame).empty ()) {
- /* We got what we want */
- break;
- }
-
- if (pass ()) {
- /* The decoder has nothing more for us */
- break;
- }
-
- if (!_decoded_video.empty() && _decoded_video.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_video (frame);
- } else {
- /* Any frame will do: use the first one that comes out of pass() */
- while (_decoded_video.empty() && !pass ()) {}
- if (!_decoded_video.empty ()) {
- dec.push_back (_decoded_video.front ());
- }
- }
-
- /* Clean up _decoded_video; keep the frame we are returning (which may have two images
- for 3D), but nothing before that */
- while (!_decoded_video.empty() && _decoded_video.front().frame < dec.front().frame) {
- _decoded_video.pop_front ();
- }
-
- return dec;
-}
-
-/** Fill _decoded_video from `from' up to, but not including, `to' */
+/** 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_2d (Frame from, Frame to)
+VideoDecoder::emit (shared_ptr<const Film> film, shared_ptr<const ImageProxy> image, Frame decoder_frame)
{
- if (to == 0) {
- /* Already OK */
+ if (ignore ()) {
return;
}
- /* Fill with black... */
- boost::shared_ptr<const ImageProxy> filler_image (new RawImageProxy (_black_image));
- Part filler_part = PART_WHOLE;
-
- /* ...unless there's some video we can fill with */
- if (!_decoded_video.empty ()) {
- filler_image = _decoded_video.back().image;
- filler_part = _decoded_video.back().part;
- }
+ auto const afr = _content->active_video_frame_rate(film);
+ auto const vft = _content->video->frame_type();
- for (Frame i = from; i < to; ++i) {
-#ifdef DCPOMATIC_DEBUG
- test_gaps++;
-#endif
- _decoded_video.push_back (
- ContentVideo (filler_image, EYES_BOTH, filler_part, i)
- );
- }
-}
+ auto frame_time = ContentTime::from_frames (decoder_frame, afr);
-/** Fill _decoded_video from `from' up to, but not including, `to' */
-void
-VideoDecoder::fill_3d (Frame from, Frame to, Eyes eye)
-{
- if (to == 0 && eye == EYES_LEFT) {
- /* Already OK */
- return;
- }
-
- /* Fill with black... */
- boost::shared_ptr<const ImageProxy> filler_left_image (new RawImageProxy (_black_image));
- boost::shared_ptr<const ImageProxy> 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<ContentVideo>::const_reverse_iterator i = _decoded_video.rbegin(); i != _decoded_video.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_video.empty() ? EYES_LEFT : _decoded_video.back().eyes;
-
- if (_decoded_video.empty ()) {
- filler_frame = 0;
- filler_eye = EYES_LEFT;
- } else if (_decoded_video.back().eyes == EYES_LEFT) {
- filler_frame = _decoded_video.back().frame;
- filler_eye = EYES_RIGHT;
- } else if (_decoded_video.back().eyes == EYES_RIGHT) {
- filler_frame = _decoded_video.back().frame + 1;
- filler_eye = EYES_LEFT;
- }
-
- while (filler_frame != to || filler_eye != eye) {
-
-#ifdef DCPOMATIC_DEBUG
- test_gaps++;
-#endif
-
- _decoded_video.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
- )
- );
-
- if (filler_eye == EYES_LEFT) {
- filler_eye = EYES_RIGHT;
+ 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.
+ */
+
+ if (vft == VideoFrameType::THREE_D_ALTERNATE) {
+ frame = decoder_frame / 2;
+ eyes = (decoder_frame % 1) ? Eyes::RIGHT : Eyes::LEFT;
} else {
- filler_eye = EYES_LEFT;
- ++filler_frame;
+ frame = decoder_frame;
+ if (vft == VideoFrameType::THREE_D) {
+ auto j2k = dynamic_pointer_cast<const J2KImageProxy>(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 subclasses when they have a video frame ready */
-void
-VideoDecoder::video (shared_ptr<const ImageProxy> image, Frame frame)
-{
- if (_ignore_video) {
- return;
+ _position = ContentTime::from_frames (frame, afr);
+ } else {
+ if (vft == VideoFrameType::THREE_D) {
+ auto j2k = dynamic_pointer_cast<const J2KImageProxy>(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;
+ }
}
- _video_content->film()->log()->log (String::compose ("VD receives %1", frame), Log::TYPE_DEBUG_DECODE);
-
- /* We may receive the same frame index twice for 3D, and we need to know
- when that happens.
- */
- bool const same = (!_decoded_video.empty() && frame == _decoded_video.back().frame);
-
- /* Work out what we are going to push into _decoded_video next */
- list<ContentVideo> to_push;
- switch (_video_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:
- to_push.push_back (ContentVideo (image, same ? EYES_RIGHT : EYES_LEFT, PART_WHOLE, frame));
+ case VideoFrameType::THREE_D_ALTERNATE:
+ {
+ 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_video
- and the things we are about to push.
- */
-
- boost::optional<Frame> from;
- boost::optional<Frame> to;
-
- if (_decoded_video.empty() && _last_seek_time && _last_seek_accurate) {
- from = _last_seek_time->frames (_video_content->video_frame_rate ());
- to = to_push.front().frame;
- } else if (!_decoded_video.empty ()) {
- from = _decoded_video.back().frame + 1;
- to = to_push.front().frame;
- }
-
- /* It has been known that this method receives frames out of order; at this
- point I'm not sure why, but we'll just ignore them.
- */
-
- if (from && to && from.get() > to.get()) {
- _video_content->film()->log()->log (
- String::compose ("Ignoring out-of-order decoded frame %1 after %2", to.get(), from.get()), Log::TYPE_WARNING
- );
- return;
- }
-
- if (from) {
- if (_video_content->video_frame_type() == VIDEO_FRAME_TYPE_2D) {
- fill_2d (from.get(), to.get ());
- } else {
- fill_3d (from.get(), to.get(), to_push.front().eyes);
- }
- }
-
- copy (to_push.begin(), to_push.end(), back_inserter (_decoded_video));
-
- /* We can't let this build up too much or we will run out of memory. We need to allow
- the most frames that can exist between blocks of sound in a multiplexed file.
- */
- DCPOMATIC_ASSERT (_decoded_video.size() <= 96);
+ _position = ContentTime::from_frames (frame, afr);
}
-void
-VideoDecoder::seek (ContentTime s, bool accurate)
-{
- _decoded_video.clear ();
- _last_seek_time = s;
- _last_seek_accurate = accurate;
-}
-/** Set this player never to produce any video data */
void
-VideoDecoder::set_ignore_video ()
+VideoDecoder::seek ()
{
- _ignore_video = true;
+ _position = boost::none;
+ _last_emitted_eyes.reset ();
+ _frame_interval_checker.reset (new FrameIntervalChecker());
}