[FIXED] How to pass video stream from one Python to another?

Issue

In my previous post, we found a way to pass an image file from one Python to another: pass video data from one python script to another

I am now trying to pass a video (successive images):

write.py

import sys
import numpy as np
import cv2
from PIL import Image
import io
import time

while True:
    img = cv2.imread('cat.jpg')
    bimg = cv2.imencode('.jpg',img)[1]
    sys.stdout.buffer.write(bimg)
    sys.stdout.flush()
    time.sleep(1)

read.py:

import sys
from PIL import Image
import io
import cv2
import numpy as np
from io import BytesIO
    
while True:
    data = sys.stdin.buffer.read()
    img_np = cv2.imdecode(np.frombuffer(BytesIO(data).read(), np.uint8), cv2.IMREAD_UNCHANGED)
    cv2.imshow('image', img_np)
    cv2.waitKey(0)

If I output the write.py data to terminal, it prints. If I manually hand data to read.py that gets read. But put them together (python3 write.py | python3 read.py) and it just hangs. write.py just writes once, and read.py never seems to get it.

My guess is that the read code is waiting for the write code to "end" before it wraps up the data package and calls it an image. Though if that were the case, I would think it that doing a flush would fix it.

Solution

I think I figured it out. In read.py, sys.stdin.buffer.read() reads and waits until the stdin pipe is closed but write.py never actually closes its stdout because of the while True loop. This proof of concept simplified example works:

write.py

import sys
import time

sys.stdout.buffer.write(b"Hello world")
sys.stdout.buffer.flush()

# Note if we comment out the code bellow it works again
while True:
    # Keep this alive but don't have `while True:pass`
    # because my computer might crash :D
    time.sleep(10)

and read.py

import sys

with open("output.txt", "w") as file:
    file.write(sys.stdin.read())

This will also hang and will never actually write anything to "output.txt". If we remove the while True loop from write.py the code will no longer hang and "Hello World" will be written to "output.py" because when write.py is finished writing it will close its process and that will close the pipe. To fix this issue I recommend changing read.py to something like this:

import sys

while True:
    with open("output.txt", "a") as file:
        file.write(sys.stdin.read(1))

Solution:

write.py

import sys
import time

MAX_FILE_SIZE = 16 # bytes

msg = b"Hello world"

# Tell `reader.py` that it needs to read x number of bytes.
length = len(msg)
# We also need to tell `read.py` how many bytes it needs to read.
# This means that we have reached the same problem as before.
# To fix that issue we are always going to send the number of bytes but
# We are going to pad it with `0`s at the start.
# https://stackoverflow.com/a/339013/11106801
length = str(length).zfill(MAX_FILE_SIZE)
sys.stdout.buffer.write(length.encode())

sys.stdout.buffer.write(msg)
sys.stdout.buffer.flush()

# We also need to tell `read.py` that it was the last file that we send
# Sending `1` means that the file has ended
sys.stdout.buffer.write(b"1")
sys.stdout.buffer.flush()

# Note if we comment out the code bellow it works again
while True:
    # Keep this alive but don't have `while True:pass`
    # because my computer might crash :D
    time.sleep(10)

and read.py

import sys
import time

MAX_FILE_SIZE = 16 # bytes

while True:
    time.sleep(1) # Make sure `write.py` has sent the data
    # Read `MAX_FILE_SIZE` number of bytes and convert it to an int
    # So that we know the size of the file comming in
    length = int(sys.stdin.buffer.read(MAX_FILE_SIZE))
    time.sleep(1) # Make sure `write.py` has sent the data

    # Here you can switch to a different file every time `writer.py`
    # Sends a new file
    with open("output.txt", "wb") as file:
        file.write(sys.stdin.buffer.read(length))

    file_ended = sys.stdin.buffer.read(1)
    if file_ended == b"1":
        # File has ended
        break
    else:
        # We are going to start reading again for the next file:
        pass

Edit: The solution works like this:

1. Send the size of the file
2. Send the actual file data
3. Send a byte that tell read.py if it should be expecting another file or not

For part 1, we just encode the length of the file as a string that is padded with 0s at the front. Note: Make sure that the MAX_FILE_SIZE is larger than the size of the largest file (large numbers will slightly decrease the performance). For part 3, if we send a "1" it will mean that there are no more files to be sent. Otherwise reader.py will wait and accept the next file. So write.py will become:

from math import log
import time
import sys
import cv2


MAX_FILE_SIZE = 62914560 # bytes
MAX_FILE_SIZE = int(log(MAX_FILE_SIZE, 2)+1)


def write_file(buffer, data, last_file=False):
   # Tell `reader.py` that it needs to read x number of bytes.
   length = len(data)
   # We also need to tell `read.py` how many bytes it needs to read.
   # This means that we have reached the same problem as before.
   # To fix that issue we are always going to send the number of bytes but
   # We are going to pad it with `0`s at the start.
   # https://stackoverflow.com/a/339013/11106801
   length = str(length).zfill(MAX_FILE_SIZE)
   with open("output.txt", "w") as file:
      file.write(length)
   buffer.write(length.encode())

   # Write the actual data
   buffer.write(data)

   # We also need to tell `read.py` that it was the last file that we send
   # Sending `1` means that the file has ended
   buffer.write(str(int(last_file)).encode())
   buffer.flush()


while True:
    img = cv2.imread("img.jpg")
    bimg = cv2.imencode(".jpg", img)[1]
    # Call write_data
    write_file(sys.stdout.buffer, bimg, last_file=False)
    # time.sleep(1) # Don't need this

and read.py will become:

from io import BytesIO
from math import log
import numpy as np
import time
import cv2
import sys


MAX_FILE_SIZE = 62914560 # bytes
MAX_FILE_SIZE = int(log(MAX_FILE_SIZE, 2)+1)


def read(buffer, number_of_bytes):
    output = b""
    while len(output) < number_of_bytes:
        output += buffer.read(number_of_bytes - len(output))
    assert len(output) == number_of_bytes, "An error occured."
    return output


def read_file(buffer):
    # Read `MAX_FILE_SIZE` number of bytes and convert it to an int
    # So that we know the size of the file comming in
    length = int(read(buffer, MAX_FILE_SIZE))

    # Here you can switch to a different file every time `writer.py`
    # Sends a new file
    data = read(buffer, length)

    # Read a byte so that we know if it is the last file
    file_ended = read(buffer, 1)

    return data, (file_ended == b"1")


while True:
    print("Reading file")
    data, last_file = read_file(sys.stdin.buffer)
    img_np = cv2.imdecode(np.frombuffer(BytesIO(data).read(), np.uint8),
                          cv2.IMREAD_UNCHANGED)
    cv2.imshow("image", img_np)
    cv2.waitKey(0)

    if last_file:
        break;

Answered By - TheLizzard

Answer Checked By - David Marino (PHPFixing Volunteer)