diff --git a/dcc/concat.py b/dcc/concat.py index 660d3d0..764f977 100644 --- a/dcc/concat.py +++ b/dcc/concat.py @@ -9,117 +9,131 @@ import numpy as np import wand.image class Concat(dcc.doom_base.Wad): - def get_parser(self, prog_name): - parser = super().get_parser(prog_name) - parser.add_argument("start_map") - parser.add_argument("end_map") - parser.add_argument("-n", "--nooverlay", action="store_true") - return parser + def get_parser(self, prog_name): + parser = super().get_parser(prog_name) + parser.add_argument("start_map") + parser.add_argument("end_map") + parser.add_argument("-n", "--nooverlay", action="store_true") + return parser - def take_action(self, parsed_args): - logging.basicConfig() - av.logging.set_level(av.logging.VERBOSE) - av.logging.restore_default_callback() - videos = self.fabricate.joinpath(parsed_args.wad).glob(f"{parsed_args.wad}_map*.mp4") - output = av.open(self.fabricate.joinpath(parsed_args.wad).joinpath(f"{parsed_args.wad}_maps{parsed_args.start_map}to{parsed_args.end_map}.mp4"), "w") - offset = 0 - # We'd like to use the concat filter here and connect everything into a - # single filter graph... but it produces a "Resource temporarily - # unavailable" error when switching to inputs after the first. Presumably - # fixable, but it's easier to just make one graph per video and mux - # everything together at the end. - for v in sorted(videos): - # TODO: Support UDoom in literally any way. - if not (v.name >= f"{parsed_args.wad}_map{parsed_args.start_map}.mp4" and - v.name <= f"{parsed_args.wad}_map{parsed_args.end_map}.mp4"): - continue + def take_action(self, parsed_args): + logging.basicConfig() + av.logging.set_level(av.logging.VERBOSE) + av.logging.restore_default_callback() + videos = self.fabricate.joinpath(parsed_args.wad).glob(f"{parsed_args.wad}_map*.mp4") + output = av.open(self.fabricate.joinpath(parsed_args.wad).joinpath(f"{parsed_args.wad}_maps{parsed_args.start_map}to{parsed_args.end_map}.mp4"), "w") + offset = 0 + # We'd like to use the concat filter here and connect everything into a + # single filter graph... but it produces a "Resource temporarily + # unavailable" error when switching to inputs after the first. Presumably + # fixable, but it's easier to just make one graph per video and mux + # everything together at the end. + for v in sorted(videos): + # TODO: Support UDoom in literally any way. + if not (v.name >= f"{parsed_args.wad}_map{parsed_args.start_map}.mp4" and + v.name <= f"{parsed_args.wad}_map{parsed_args.end_map}.mp4"): + continue - chunk = av.open(v) - ograph = av.filter.Graph() - sink = ograph.add("buffersink") - asink = ograph.add("abuffersink") + chunk = av.open(v) + if not (len(chunk.streams.video) == 1 and len(chunk.streams.audio) == 1): + raise Exception(f"irregular chunk {v}: streams {chunk.streams} (expected 1 video & 1 audio)") - if not parsed_args.nooverlay: - img = wand.image.Image(height=chunk.streams[0].height,width=chunk.streams[0].width) - mapstring = v.name[-6:-4] - text = self._config["map_names"][f"map{mapstring}"] - dcc.text.draw_text(img, f"MAP{mapstring}: {text}", font_size=120) - img.trim(reset_coords=True) - img.border("graya(25%, 25%)", 10, 10) - img.border(dcc.config.TEXT_STROKE_COLOR, 16, 16) - # for this to work... the image needs to have a width that's a multiple - # of 8. dude whyyyyyyy - padfactor=8 - img.border("transparent", padfactor, 0) - img.crop(width=img.width-img.width%padfactor, height=img.height) - text_frame = av.video.frame.VideoFrame(img.width, img.height, format="rgba") + ograph = av.filter.Graph() + sink = ograph.add("buffersink") + asink = ograph.add("abuffersink") - if len(output.streams.get()) == 0: - # TODO: less hardcoding. - output.add_stream("h264", rate=61440) - output.streams[0].extradata = copy.deepcopy(chunk.streams[0].extradata) - output.streams[0].height=1440 - output.streams[0].width=2560 - output.streams[0].profile="High" - output.streams[0].qmax = chunk.streams[0].qmax - output.streams[0].qmin = chunk.streams[0].qmin - output.streams[0].codec_context.gop_size=30 - output.streams[0].codec_context.max_b_frames=2 - output.streams[0].codec_context.framerate = fractions.Fraction(60,1) - output.streams[0].codec_context.pix_fmt="yuv420p" - output.streams[0].codec_context.bit_rate = chunk.streams[0].codec_context.bit_rate - output.add_stream("aac", rate=48000) - output.streams[1].extradata = copy.deepcopy(output.streams[1].extradata) - output.streams[1].rate=48000 - output.streams[1].bit_rate=chunk.streams[1].bit_rate - src = ograph.add_buffer(template=chunk.streams[0], time_base=chunk.streams[0].time_base) - asrc = ograph.add_abuffer(template=chunk.streams[1], time_base=chunk.streams[1].time_base) - ifade = ograph.add("fade", args="in:0:60") - iafade = ograph.add("afade", args="in:{}:48000".format(offset*48000/1000000)) - ofade = ograph.add("fade", args="out:{}:60".format((chunk.duration*60/1000000)-60)) - oafade = ograph.add("afade", args="out:{}:48000".format(((offset+chunk.duration)*48000/1000000)-48000)) - if not parsed_args.nooverlay: - overlay = ograph.add_buffer(width=img.width, height=img.height, format="rgba", time_base=chunk.streams[0].time_base) - overlay_fo = ograph.add("fade", args="out:240:60") - overlay.link_to(overlay_fo, 0, 0) - composite = ograph.add("overlay", args="x=4:y=4") - src.link_to(composite, 0, 0) - overlay_fo.link_to(composite, 0, 1) - composite.link_to(ifade, 0, 0) - else: - src.link_to(ifade, 0, 0) - asrc.link_to(iafade, 0, 0) - ifade.link_to(ofade, 0, 0) - iafade.link_to(oafade, 0, 0) - ofade.link_to(sink, 0, 0) - oafade.link_to(asink, 0, 0) + if not parsed_args.nooverlay: + img = wand.image.Image(height=chunk.streams[0].height,width=chunk.streams[0].width) + mapstring = v.name[-6:-4] + text = self._config["map_names"][f"map{mapstring}"] + dcc.text.draw_text(img, f"MAP{mapstring}: {text}", font_size=120) + img.trim(reset_coords=True) + img.border("graya(25%, 25%)", 10, 10) + img.border(dcc.config.TEXT_STROKE_COLOR, 16, 16) + # for this to work... the image needs to have a width that's a multiple + # of 8. dude whyyyyyyy + padfactor=8 + img.border("transparent", padfactor, 0) + img.crop(width=img.width-img.width%padfactor, height=img.height) - ograph.configure() - for packet in chunk.demux(): - if packet.dts is None: - continue - packet.dts += (offset * packet.time_base.denominator) / (packet.time_base.numerator * 1000000) - packet.pts += (offset * packet.time_base.denominator) / (packet.time_base.numerator * 1000000) - if packet.stream_index == 0: # TODO: robustness - for ifr in packet.decode(): - if not parsed_args.nooverlay: - text_frame = av.video.frame.VideoFrame(img.width, img.height, format="rgba") - text_frame.planes[0].update(img.make_blob(format="rgba")) - text_frame.pts = ifr.pts - text_frame.dts = ifr.dts - text_frame.time_base = ifr.time_base - overlay.push(text_frame) - src.push(ifr) - ofr = sink.pull() - for p in output.streams[packet.stream_index].encode(ofr): - output.mux(p) - else: - for ifr in packet.decode(): - asrc.push(ifr) - ofr = asink.pull() - for p in output.streams[packet.stream_index].encode(ofr): - output.mux(p) - offset += chunk.duration - chunk.close() - output.close() + if len(output.streams.get()) == 0: + # We can't use the input stream as a template here; it doesn't + # have everything needed to do encoding and will fail + # mysteriously later. + vs = chunk.streams.video[0] + output.add_stream("h264", rate=int(vs.time_base.denominator/vs.time_base.numerator)) + output.streams[0].extradata = copy.deepcopy(vs.extradata) + output.streams[0].height=vs.height + output.streams[0].width=vs.width + output.streams[0].qmax = vs.qmax + output.streams[0].qmin = vs.qmin + output.streams[0].codec_context.bit_rate = vs.codec_context.bit_rate + output.streams[0].codec_context.framerate = vs.base_rate + output.streams[0].codec_context.pix_fmt = vs.codec_context.pix_fmt + # The following are only used for encoding and have no equivalent on the input stream. + output.streams[0].profile="High" + output.streams[0].codec_context.gop_size=30 + output.streams[0].codec_context.max_b_frames=2 + + astr = chunk.streams.audio[0] + output.add_stream("aac", rate=astr.rate) + output.streams[1].extradata = copy.deepcopy(astr.extradata) + output.streams[1].bit_rate=astr.bit_rate + + src = ograph.add_buffer(template=chunk.streams.video[0], time_base=chunk.streams.video[0].time_base) + asrc = ograph.add_abuffer(template=chunk.streams.audio[0], time_base=chunk.streams.audio[0].time_base) + # TODO: video fades are absolute relative to the input video; audio + # fades need to have their timestamps offset by the position in the + # final video. Clarify if this is really necessary. + frame_rate = chunk.streams.video[0].base_rate + sample_rate = chunk.streams.audio[0].rate + ifade = ograph.add("fade", args="in:0:{}".format(frame_rate)) + ofade = ograph.add("fade", args="out:{}:{}".format((chunk.duration*frame_rate/1000000)-frame_rate, frame_rate)) + iafade = ograph.add("afade", args="in:{}:{}".format(offset*sample_rate/1000000, sample_rate)) + oafade = ograph.add("afade", args="out:{}:{}".format(((offset+chunk.duration)*sample_rate/1000000)-sample_rate, sample_rate)) + if not parsed_args.nooverlay: + overlay = ograph.add_buffer(width=img.width, height=img.height, format="rgba", time_base=chunk.streams[0].time_base) + overlay_fo = ograph.add("fade", args="out:{}:{}".format(4*frame_rate, frame_rate)) + overlay.link_to(overlay_fo, 0, 0) + composite = ograph.add("overlay", args="x=4:y=4") + src.link_to(composite, 0, 0) + overlay_fo.link_to(composite, 0, 1) + composite.link_to(ifade, 0, 0) + else: + src.link_to(ifade, 0, 0) + + asrc.link_to(iafade, 0, 0) + ifade.link_to(ofade, 0, 0) + iafade.link_to(oafade, 0, 0) + ofade.link_to(sink, 0, 0) + oafade.link_to(asink, 0, 0) + ograph.configure() + + for packet in chunk.demux(): + if packet.dts is None: + continue + packet.dts += (offset * packet.time_base.denominator) / (packet.time_base.numerator * 1000000) + packet.pts += (offset * packet.time_base.denominator) / (packet.time_base.numerator * 1000000) + if packet.stream == chunk.streams.video[0]: + for ifr in packet.decode(): + if not parsed_args.nooverlay: + text_frame = av.video.frame.VideoFrame(img.width, img.height, format="rgba") + text_frame.planes[0].update(img.make_blob(format="rgba")) + text_frame.pts = ifr.pts + text_frame.dts = ifr.dts + text_frame.time_base = ifr.time_base + overlay.push(text_frame) + src.push(ifr) + ofr = sink.pull() + for p in output.streams[packet.stream_index].encode(ofr): + output.mux(p) + else: + for ifr in packet.decode(): + asrc.push(ifr) + ofr = asink.pull() + for p in output.streams[packet.stream_index].encode(ofr): + output.mux(p) + offset += chunk.duration + chunk.close() + output.close()