Switch to gstreamer for video encoding

This will enable better progress indication, and on-line recording in
future.
This commit is contained in:
Pavel Machek
2024-05-05 19:46:19 +02:00
parent 950d11fd96
commit a3dea3503d
2 changed files with 363 additions and 131 deletions

View File

@@ -28,15 +28,9 @@ elif [ "-$1" == "-stop" ]; then
mkdir $GIGA_DIR/sm
kill `cat $2/audio.pid`
cd $GIGA_DIR
echo Message: Jp
@LIBEXECDIR@/mpegize.py jpegize $GIGA_DIR/ $FPS
echo Message: Mp
cd $GIGA_DIR/sm
@LIBEXECDIR@/mpegize.py convertall $GIGA_DIR/ $FPS
echo Message: Cl
mv $GIGA_DIR/smo/*.mp4 $DEST_NAME
@LIBEXECDIR@/mpegize.py convert $GIGA_DIR/ $DEST_NAME $FPS
rm -r $GIGA_DIR
echo Message: Ok
echo Message: Rec
else
echo "Unrecognized command"
fi

View File

@@ -2,6 +2,11 @@
# Copyright 2022, 2024 Pavel Machek, GPLv2+
import os, sys, time, copy, subprocess
import gi
gi.require_version('Gst', '1.0')
from gi.repository import Gst, GLib
import os
import time
# https://stackoverflow.com/questions/11779490/how-to-add-a-new-audio-not-mixing-into-a-video-using-ffmpeg
# https://ottverse.com/create-video-from-images-using-ffmpeg/
@@ -13,6 +18,349 @@ import os, sys, time, copy, subprocess
# Usage: mpegize convert
# head -c 1000 < /dev/zero > /tmp/delme.sm/1.foo.sa
def gst_convert(mega_dir, out_file, use_jpeg):
def sa_read(name, t):
with open(name, "rb") as file:
rgb_data = file.read(10*1024*1024)
caps_string = "audio/x-raw,format=U16LE,channels=2,rate=48000,layout=interleaved,channel-mask=3"
caps = Gst.Caps.from_string(caps_string)
buffer = Gst.Buffer.new_wrapped(rgb_data)
if False:
time.sleep(1/30.)
# nanoseconds
buffer.pts = time.time() * 1000*1000*1000
buffer.dts = time.time() * 1000*1000*1000
elif True:
buffer.pts = t
buffer.dts = t
buffer.duration = (1000*1000*1000)/10.
return buffer, caps
def sa_src(appsrc):
def on_need_data(appsrc, data, length):
name = audio.get_path()
if name == None or name[-22:] != ".48000-s16le-stereo.sa":
appsrc.emit("end-of-stream")
print("End of audio stream")
return
t = audio.get_time()
#print("Audio: ", name, " need ", data, t)
audio.pop()
buffer, caps = sa_read(name, t)
appsrc.set_property("caps", caps)
appsrc.emit("push-buffer", buffer)
appsrc.set_property("format", Gst.Format.TIME)
appsrc.set_property("is-live", False)
appsrc.set_property("block", True)
name = audio.get_path()
buffer, caps = sa_read(name, 0)
appsrc.set_property("caps", caps)
#appsrc.emit("push-buffer", buffer)
s = appsrc.connect("need-data", on_need_data, "")
print("Connect", s)
class grwBase:
def init(m, dir):
m.dir = dir
m.list = os.listdir(dir)
m.list.sort()
m.slen = len(m.suffix)
m.start_time = 0
print("Movie", len(m.list))
def get_path(m):
s = m.get_name()
if s: return m.dir + s
return s
def get_name(m):
#print("Get path -- ")
while True:
if (len(m.list)) == 0:
return None
#print("Get path: ", m.list[0], m.suffix)
if m.list[0][-m.slen:] != m.suffix:
m.pop()
continue
return m.list[0]
def get_time(m):
s = m.get_name()
s = s[:-m.slen]
return int(s) * 1000 - m.start_time
def pop(m):
m.list = m.list[1:]
class grwVideo(grwBase):
suffix = ".grw"
def __init__(m, dir):
m.init(dir)
class grwJPEG(grwBase):
suffix = ".jpeg.sv"
def __init__(m, dir):
m.init(dir + "sm/")
class grwAudio(grwVideo):
suffix = ".48000-s16le-stereo.sa"
def __init__(m, dir):
m.init(dir + "sm/")
def grw_read(name, t):
with open(name, "rb") as file:
rgb_data = file.read(10*1024*1024)
i = len(rgb_data)
i -= 1
while rgb_data[i] != 0:
i -= 1
footer = rgb_data[i+1:]
sp = str(footer, 'ascii').split('\n')
# Create caps for the file
caps_string = sp[0][6:]
caps = Gst.Caps.from_string(caps_string)
if sp[0][:6] != "Caps: ":
print("Bad footer")
if sp[1][:6] != "Size: ":
print("Bad footer")
if sp[-1] != "GRW":
print("Missing GRW footer")
buffer = Gst.Buffer.new_wrapped(rgb_data)
# This does not work for interactive use.
if False:
time.sleep(1/30.)
# nanoseconds
buffer.pts = time.time() * 1000*1000*1000
buffer.dts = time.time() * 1000*1000*1000
elif True:
buffer.pts = t
buffer.dts = t
buffer.duration = (1000*1000*1000)/30.
return buffer, caps
def grwsrc(appsrc):
def on_need_data(appsrc, data, length):
name = movie.get_path()
if name == None or name[-4:] != ".grw":
appsrc.emit("end-of-stream")
print("End of video stream")
return
t = movie.get_time()
#print("Video: ", name, t)
movie.pop()
buffer, caps = grw_read(name, t)
appsrc.set_property("caps", caps)
appsrc.emit("push-buffer", buffer)
appsrc.set_property("format", Gst.Format.TIME)
appsrc.set_property("is-live", False)
appsrc.set_property("block", True)
name = movie.get_path()
buffer, caps = grw_read(name, 0)
appsrc.set_property("caps", caps)
#appsrc.emit("push-buffer", buffer)
s = appsrc.connect("need-data", on_need_data, "")
print("Connect", s)
def jpeg_read(name, t):
with open(name, "rb") as file:
rgb_data = file.read(10*1024*1024)
i = len(rgb_data)
buffer = Gst.Buffer.new_wrapped(rgb_data)
caps_string = "image/jpeg"
caps = Gst.Caps.from_string(caps_string)
# This does not work for interactive use.
if False:
time.sleep(1/30.)
# nanoseconds
buffer.pts = time.time() * 1000*1000*1000
buffer.dts = time.time() * 1000*1000*1000
elif True:
buffer.pts = t
buffer.dts = t
buffer.duration = (1000*1000*1000)/30.
return buffer, caps
def jpeg_src(appsrc):
def on_need_data(appsrc, data, length):
name = movie.get_path()
if name == None or name[-8:] != ".jpeg.sv":
appsrc.emit("end-of-stream")
print("End of video stream")
return
t = movie.get_time()
#print("Video: ", name, t)
movie.pop()
buffer, caps = jpeg_read(name, t)
appsrc.set_property("caps", caps)
appsrc.emit("push-buffer", buffer)
appsrc.set_property("format", Gst.Format.TIME)
appsrc.set_property("is-live", False)
appsrc.set_property("block", True)
name = movie.get_path()
buffer, caps = jpeg_read(name, 0)
appsrc.set_property("caps", caps)
#appsrc.emit("push-buffer", buffer)
s = appsrc.connect("need-data", on_need_data, "")
print("Connect", s)
def v_src(appsrc):
if not use_jpeg:
grwsrc(appsrc)
else:
jpeg_src(appsrc)
count = 0
path = mega_dir
if use_jpeg:
movie = grwJPEG(path)
else:
movie = grwVideo(path)
audio = grwAudio(path)
t1 = movie.get_time()
t2 = audio.get_time()
tm = min(t1,t2)
print("Time base is", tm)
movie.start_time = tm
audio.start_time = tm
def pipeline_video():
if True:
s = "appsrc name=source"
if use_jpeg:
s += " ! jpegdec "
else:
s = "videotestsrc"
s += " ! video/x-raw,width=(int)640,height=(int)480,format=(string)RGB "
if False:
s += " ! videoconvert ! jpegenc"
s += " ! appsink name=sink"
elif True:
s += " ! videoconvert ! autovideosink"
else:
s += " ! videoconvert ! x264enc bitrate=3072 speed-preset=ultrafast ! matroskamux ! filesink location=" + out_file
pipeline = Gst.parse_launch(s)
p = pipeline.get_by_name("source")
if p:
if False:
mysrc(p)
else:
v_src(p)
p = pipeline.get_by_name("sink")
if p:
mysink(p)
return pipeline
def pipeline_audio():
# audiotestsrc ! audioconvert ! audioresample ! autoaudiosink
if True:
s = "appsrc name=source"
else:
s = "audiotestsrc"
if True:
s += " ! audiobuffersplit ! audioconvert ! audioresample ! autoaudiosink"
else:
s += " ! ! ! "
pipeline = Gst.parse_launch(s)
p = pipeline.get_by_name("source")
if p:
sa_src(p)
p = pipeline.get_by_name("sink")
if p:
mysink(p)
return pipeline
def pipeline_both():
if True:
s = "appsrc name=asrc"
else:
s = "audiotestsrc"
# Audiobuffersplit creates problems with A/V synchronization, avoid.
#s += "! audiobuffersplit"
s += " ! audioconvert ! vorbisenc ! mux. "
if True:
s += "appsrc name=vsrc"
if use_jpeg:
s += " ! jpegdec "
else:
s += "videotestsrc"
s += " ! video/x-raw,width=(int)640,height=(int)480,format=(string)RGB "
s += " ! videoconvert ! x264enc bitrate=3072 speed-preset=ultrafast ! matroskamux name=mux"
if False:
s += " ! decodebin ! playsink"
else:
s += " ! filesink location="+out_file
pipeline = Gst.parse_launch(s)
p = pipeline.get_by_name("asrc")
if p:
sa_src(p)
p = pipeline.get_by_name("vsrc")
if p:
v_src(p)
return pipeline
Gst.init(None)
Gst.debug_set_default_threshold(Gst.DebugLevel.WARNING)
if False:
Gst.debug_set_default_threshold(Gst.DebugLevel.INFO)
if False:
pipeline = pipeline_video()
elif False:
pipeline = pipeline_audio()
else:
pipeline = pipeline_both()
# Function to handle end of stream
def on_eos(bus, message):
print("End of stream")
pipeline.set_state(Gst.State.NULL)
loop.quit()
# Set up bus to handle messages
bus = pipeline.get_bus()
bus.add_signal_watch()
bus.connect("message::eos", on_eos)
# Set the pipeline to the playing state
pipeline.set_state(Gst.State.PLAYING)
# Run the main loop to handle GStreamer events
loop = GLib.MainLoop()
try:
loop.run()
except KeyboardInterrupt:
pipeline.set_state(Gst.State.NULL)
loop.quit()
class Mpegize:
base = '/tmp/delme.'
fps = 30.5
@@ -48,14 +396,22 @@ class Mpegize:
return s[2], s[1], i
def help(m):
print("mpegize command base-dir")
print("mpegize command base-dir destination-movie fps")
def run(m, argv):
if len(argv) > 2:
m.base = argv[2]
mode = argv[1]
if mode == "stat" or mode == "convert" or mode == "gc" or mode == "convertall":
m.process(mode)
if mode == "convert":
print("Phase 1: jpegize")
print("Message: 0%%")
sys.stdout.flush()
m.prepare()
m.jpegize()
print("Phase 2: mpegize -- ", argv[3])
print("Message: enc")
sys.stdout.flush()
gst_convert(m.base, argv[3], True)
return
if mode == "gaps":
print("Video gaps")
@@ -87,131 +443,12 @@ class Mpegize:
print("Total", num, "gaps of", total)
print("Expected", (1000000 / m.fps) / 1000., "msec, limit", limit / 1000., "msec")
def process(m, mode):
m.prepare_source()
photos = 0
video_frames = 0
start = 0
for n in m.frames:
ext, mid, i = m.parse_frame(n)
if ext != "mark":
continue
print(n)
if mid == "start":
start = i
if mid == "stop":
video_frames += m.extract_video(start, i, mode)
start = 0
if mid == "wow":
if start:
start -= 5000000
else:
photos += 5
m.extract_photo(i - 1000000, mode)
m.extract_photo(i - 2000000, mode)
m.extract_photo(i - 3000000, mode)
m.extract_photo(i - 4000000, mode)
m.extract_photo(i - 5000000, mode)
if mid == "photo":
photos += 1
m.extract_photo(i, mode)
if mode == "convertall":
video_frames += m.extract_video(0, 9999999999999999, "convert")
return
print("Total", photos, "photos and", m.display_frames(video_frames))
print(len(m.unused_frames), "/", len(m.frames))
if mode == "gc":
os.chdir(m.source)
for n in m.unused_frames:
os.unlink(n)
print(m.unused_frames)
def display_usec(m, v):
return "%.2f sec" % (v/1000000.)
def display_frames(m, v):
return "%d frames %s" % (v, m.display_usec(v * 1000000 / 30.))
def frame_used(m, n):
if n in m.unused_frames:
m.unused_frames.remove(n)
def extract_photo(m, around, mode):
print("Photo:", around)
best = None
for n in m.frames:
ext, mid, i = m.parse_frame(n)
if ext != "sv":
continue
if i < around:
best = n
continue
best = n
break
m.frame_used(best)
out_file = m.output+"/image-%04d.jpeg" % m.out_index
m.out_index += 1
if mode == "convert":
os.system("ln "+m.source+"/"+best+" "+out_file)
def extract_video(m, start, end, mode):
print("Searching video", start, end, "--", m.display_usec(end-start))
if mode == "convert":
m.prepare_work()
t1 = time.time()
seen_audio = False
seen_video = False
count = 0
skip_audio = 0
skip_video = 0
num = 0
for n in m.frames:
num += 1
if not num % 1000:
print("Frame", num)
ext, mid, i = m.parse_frame(n)
if ext != "sa" and ext != "sv":
m.frame_used(n)
continue
if i < start - 1000000 or i > end:
continue
if ext == "sa":
seen_audio = True
if not seen_video:
continue
if mode == "convert":
os.system("cat "+m.source+"/"+n+" >> "+m.work+"/output.raw")
if ext == "sv":
if not seen_video:
first_video = i
seen_video = True
if mode == "convert":
os.system("ln "+m.source+"/"+n+" "+m.work+"/image-%06d.jpeg" % count)
count += 1
while i >= first_video + count * 1000000 / m.fps:
print("Duplicating video frame at", i)
if mode == "convert":
os.system("ln "+m.source+"/"+n+" "+m.work+"/image-%06d.jpeg" % count)
count += 1
m.frame_used(n)
if mode == "convert":
os.chdir(m.work)
print("Converting", m.display_frames(count), "skipped", skip_audio, "audio and", skip_video, "video frames")
os.system("ffmpeg -f s16le -ac 2 -ar 48000 -i output.raw output.wav")
options = "-b:v 4096k -c:v libx264 -preset ultrafast"
os.system("ffmpeg -framerate %d -i image-%%06d.jpeg -i output.wav %s output.mp4" % (m.fps, options))
os.system("rm output.raw")
out_file = m.output+"/video-%04d.mp4" % m.out_index
m.out_index += 1
os.system("mv output.mp4 "+out_file)
print("Converted", m.display_frames(count), "in", "%.1f" % (time.time()-t1), "seconds")
if mode == "convert":
print("Original size -> new size")
os.system("du -sh .; du -sh "+out_file)
return count
def jpegize(m):
i = 0
os.chdir(m.base)
@@ -225,7 +462,8 @@ class Mpegize:
print("Something went terribly wrong")
continue
i += 1
print(i, '/', len(l))
print("Message: %.0f%%" % ((100*i) / len(l)))
sys.stdout.flush()
base = n[:-4]
subprocess.run(['dcraw',
'-w', # -w Use camera white balance