Memory issues when loading videos into frames

asked 2018-09-13 06:59:32 -0600

Make42 gravatar image

updated 2018-09-13 07:41:05 -0600

I have folder with 160 FLV videos, each having 120 frames of size 152, 360 with RGB colors (3 channels) that I would like to load into the numpy array frames. I do this with the code:

import numpy as np
import cv2
import os

directory = "data/"
# frames = []
frames = np.empty(shape=(160 * 120, 152, 360, 3), dtype=np.float32)

for file in os.listdir(directory):
    if file.endswith(".flv"):
        file_path = os.path.join(directory, file)
        nr_file = nr_file + 1
        print('File '+str(nr_file)+' of '+str(nb_files_in_dir)+' files: '+file_path)

        # Create a VideoCapture object and read from input file
        # If the input is the camera, pass 0 instead of the video file name
        cap = cv2.VideoCapture(file_path)

        # Check if camera opened successfully
        if (cap.isOpened() == False):
            print("Error opening video stream or file")

        # Read until video is completed
        while (cap.isOpened()):
            # Capture frame-by-frame
            ret, frame =
            if ret == True:
                # frames.append(frame.astype('float32') / 255.)
                frames[nr_frame, :, :, :] = frame.astype('float32') / 255.
                nr_frame = nr_frame + 1
                nb_frames_in_file = nb_frames_in_file + 1

        # When everything done, release the video capture object

# frames = np.array(frames)

Originally I tried to use a list frames (see the commented lines), instead of the prerallocated numpy array, but it seemed this took too much memory - no idea why though.

However, it seems this did not help much: Still the code is very memory hungry (many GB), even though my videos are just a few KB large. I think it is because the resources of the cap-objects (the cv2.VideoCapture-objects) might not freed despite me using cap.release() - is that correct? What can I do, to make my code memory-efficient?

edit retag flag offensive close merge delete


no, it's not the videocapture, the decompressed frames just need a huge amount of memory.

just do the maths. it is:

320*240*3 = 216000 bytes for a single image,
120*216000 = 25920000 bytes per movie and 
25920000*1000 = 25920000000 bytes for 1000 movies :\

you'll have to restrict it somehow ...

berak gravatar imageberak ( 2018-09-13 07:11:16 -0600 )edit

oh, you also convert to float, so the whole thing * 4

(why do you think, that's nessecary ?)

berak gravatar imageberak ( 2018-09-13 07:23:52 -0600 )edit

@berak: I corrected my mistake: There are only 160 images and they are smaller afterall. If the size of data itself would be the issue, then the allocation of the frames numpy array would already eat up all memory, but it does not. The float is because I am putting the thing into a neural network afterwards. The numpy array is actually not that large, even if it is allocated with float32, so this should not be the issue (I think).

Make42 gravatar imageMake42 ( 2018-09-13 07:34:28 -0600 )edit

still, you're trying to allocate ~25gb of memory for this.

you'll have to feed it into the nn in batches later, so only load 1 batch at a time.

berak gravatar imageberak ( 2018-09-13 08:14:12 -0600 )edit

Thanks, that is what I am doing now. Since I need a DataGenerator I implemented a keras.utils.Sequence class and use this for batch-training of my neural network.

Make42 gravatar imageMake42 ( 2018-09-18 03:49:32 -0600 )edit