Fast Video Processing in Python

• Fast video processing in python using batch-processing and multi-processing.
  • Fast reading a video
  • Fast writing a video
  • Fast saving frames in a video

Table of contents

1. Notice
2. License
3. Development envrionment
4. Preparation to run a demo code
5. Run a demo code
6. Experiment result
7. Revision history

1. Notice

• I recommend that you should ignore the commented instructions with an octothorpe, #.
• Modified date: May 31, 2020.
• Developed version: 200531a

2. License

• All providen codes comply with the MIT license.

3. Development environment

• Operating System (OS): Ubuntu MATE 18.04.3 LTS (Bionic)
• CPU: Intel Core i7-7700 CPU @ 3.60GHz x 8
• Main memory: 32GB
• GPU: Titan XP, 1ea
• Python version: Python 3.7.6
• Python main package:
  • opencv-python: 4.1.2.30
  • ffmpeg-python: 0.2.0

4. Preparation to run a demo code

A. Reference to the website, virtualenv-python, opencv-python and ffmpeg-python.

• If you don't have installed Python, please refer to the site, virtualenv-python, above.

B. Activate a virtualenv.

• The root directory for the virtualenv: /home/usrname/pip3_virtualenv
• The name of virtualenv to be activated: virenv_dl

usrname@hostname:~/curr_path$ source /home/usrname/pip3_virtualenv/virenv_dl/bin/activate

C. Install the required main packages.

(virenv_dl) usrname@hostname:~/curr_path$ pip3 install opencv-python
(virenv_dl) usrname@hostname:~/curr_path$ pip3 install ffmpeg-python

5. Run a demo code

• A demo code is a test code that reads a video and saves all frames in the video or writes another video.
• The measured time is the time including all the prcoessing time (e.g. reading, saving or writing).
• I use a HD1080 video (i.e. ./test_input/test_1.mp4) as a test video source.
• Please note that I cannot provide a test video, but you can test any video you want.
• Please note that you should check the arguements in the main python script. Some examples are as follows.
  • name_video_src: The name of test video source.
  • num_batch: The numebr of batch size for batch-processing in reading a video.
  • num_proc: The numebr of processes for multi-processing in wrting frames.

A. Reading a vdieo: opencv-python, Saving frames: opencv-python (single-process)

(virenv_dl) usrname@hostname:~/curr_path$ python3 main_saving_frames.py --opt_reading_opencv True --opt_writing_single_proc True

B. Reading a vdieo: opencv-python, Saving frames: opencv-python (multi-process)

(virenv_dl) usrname@hostname:~/curr_path$ python3 main_saving_frames.py --opt_reading_opencv True --opt_writing_single_proc False

C. Reading a vdieo: ffmpeg-python, Saving frames: opencv-python (single-process)

(virenv_dl) usrname@hostname:~/curr_path$ python3 main_saving_frames.py --opt_reading_opencv False --opt_writing_single_proc True

D. Reading a vdieo: ffmpeg-python, Saving frames: opencv-python (multi-process)

(virenv_dl) usrname@hostname:~/curr_path$ python3 main_saving_frames.py --opt_reading_opencv False --opt_writing_single_proc False

E. Reading a vdieo: opencv-python, Writing a video: opencv-python

(virenv_dl) usrname@hostname:~/curr_path$ python3 main_writing_video.py --opt_reading_opencv True --opt_writing_opencv True

F. Reading a vdieo: opencv-python, Writing a video: ffmpeg-python

(virenv_dl) usrname@hostname:~/curr_path$ python3 main_writing_video.py --opt_reading_opencv True --opt_writing_opencv False

G. Reading a vdieo: ffmpeg-python, Writing a video: opencv-python

(virenv_dl) usrname@hostname:~/curr_path$ python3 main_writing_video.py --opt_reading_opencv False --opt_writing_opencv True

H. Reading a vdieo: ffmpeg-python, Writing a video: ffmpeg-python

(virenv_dl) usrname@hostname:~/curr_path$ python3 main_writing_video.py --opt_reading_opencv False --opt_writing_opencv False

6. Experiment result

No.	Description	Total time [Sec.]	Total FPS	Averaged time [Sec.]	Averaged FPS
1	Reading a video with opencv-python	1.94	0.52	0.01	188.42
2	Reading a video with ffmpeg-python	2.59	0.39	0.01	140.84
3	Saving all frames with opencv-python (single-process)	17.77	0.06	0.05	20.54
4	Saving all frames with opencv-python (multi-process)	0.01	89.66	3.1e-5	32726.30
5	Writing a video with opencv-python	5.44	0.18	0.01	67.15
6	Writing a video with ffmpeg-python	3.27	0.31	0.01	111.65

• As far as I mentioned in section 5, the measured time is the time including all the prcoessing time (e.g. reading, saving or writing).
• In the table, the meaning of the word in the cell, averaged, is the average processing time per frame.
• The test envrionment including main arguemnts are as follows:
  • The resolution of the video: HD1080
  • The FPS of the video: about 23 FPS
  • The duration of the video: 15 seconds
  • The number of total frames in the video: 361
  • The unit for batch-processing (i.e. num_batch): 5
  • The unit for multi-processing (i.e. num_proc): 4

7. Revision history

Version	Description	Modifier
200531a	Release the code for the first time.	vujadeyoon