Can OpenCV's VideoWriter write in a separate process? - python-3.x

Question

I'm trying to save a video to disk in a separate process. The program creates a buffer of images to save on the original process. When its done recording, it passes the file name and image buffer to a second process that will make its own VideoWriter and save the file. When the second process calls write, however, nothing happens. It hangs and doesn't output any errors.
I checked if the VideoWriter is open already and it is. I tried moving the code to the original process to see if it worked there and it does. I don't know if it is some setting I need to initialize in the new process or if it has to do with the way VideoWriter works.
Here's my code
def stop_recording(self):
"""Stops recording in a separate process"""
if self._file_dump_process is None:
self._parent_conn, child_conn = multiprocessing.Pipe()
self._file_dump_process = multiprocessing.Process(
target=self.file_dump_loop, args=(child_conn, self.__log))
self._file_dump_process.daemon = True
self._file_dump_process.start()
if self._recording:
self.__log.info("Stopping recording. Please wait...")
# Dump VideoWriter and image buffer to process
# Comment out when running on main procress
self._parent_conn.send([self._record_filename, self._img_buffer])
""" Comment in when running on main procress
fourcc = cv2.VideoWriter_fourcc(*"MJPG")
effective_fps = 16.0
frame_shape = (640, 480)
record_file = cv2.VideoWriter(self._record_filename, fourcc,
effective_fps, frame_shape,
isColor=1)
for img in self._img_buffer:
self.__log.info("...still here...")
record_file.write(img)
# Close the file and set it to None
record_file.release()
self.__log.info("done.")
"""
# Delete the entire image buffer no matter what
del self._img_buffer[:]
self._recording = False
#staticmethod
def file_dump_loop(child_conn, parent_log):
fourcc = cv2.VideoWriter_fourcc(*"MJPG")
effective_fps = 16.0
frame_shape = (640, 480)
while True:
msg = child_conn.recv()
record_filename = msg[0]
img_buffer = msg[1]
record_file = cv2.VideoWriter(record_filename, fourcc,
effective_fps, frame_shape,
isColor=1)
for img in img_buffer:
parent_log.info("...still here...")
record_file.write(img)
# Close the file and set it to None
record_file.release()
del img_buffer[:]
parent_log.info("done.")
Here's the log output when I run it on one process:
2019-03-29 16:19:02,469 - image_processor.stop_recording - INFO: Stopping recording. Please wait...
2019-03-29 16:19:02,473 - image_processor.stop_recording - INFO: ...still here...
2019-03-29 16:19:02,515 - image_processor.stop_recording - INFO: ...still here...
2019-03-29 16:19:02,541 - image_processor.stop_recording - INFO: ...still here...
2019-03-29 16:19:02,567 - image_processor.stop_recording - INFO: ...still here...
2019-03-29 16:19:02,592 - image_processor.stop_recording - INFO: ...still here...
2019-03-29 16:19:02,617 - image_processor.stop_recording - INFO: ...still here...
2019-03-29 16:19:02,642 - image_processor.stop_recording - INFO: ...still here...
2019-03-29 16:19:02,670 - image_processor.stop_recording - INFO: done.
Here's the log output when I run it on a second process:
2019-03-29 16:17:27,299 - image_processor.stop_recording - INFO: Stopping recording. Please wait...
2019-03-29 16:17:27,534 - image_processor.file_dump_loop - INFO: ...still here...

Answer 1

I tried this, and was successful with the following code:
import cv2
cap, imgs = cv2.VideoCapture('exampleVideo.MP4'), []
# This function writes video
def write_video(list_of_images):
vid_writer = cv2.VideoWriter('/home/stephen/Desktop/re_encode.avi',cv2.VideoWriter_fourcc('M','J','P','G'),120, (640,480))
for image in list_of_images: vid_writer.write(image)
# Loop to read video and save images to a list
for frame in range(123):
_, img = cap.read()
imgs.append(img)
write_video(imgs)
cap.release()
Everything worked as expected, and when I checked how long it took to run, I found that the above code took .13 seconds to read the video and .43 seconds to write the video. If I read the video and write the video in the same loop (below) the total processing time is .56 seconds (which is .13+.43).
# Loop to save image to video
for frame in range(123):
_, img = cap.read()
vid_writer.write(img)
There is a big disadvantage of writing the images to a buffer (in memory) first, and then writing the images to a video file (on the hard drive). The buffer is saved on the RAM which will fill up very quickly, and you will likely get a memory error.