videoio: save/load image sequence as H.264 video

A small Python module for saving and loading RGB and uint16 (depth) frames as H.264 encoded video

Prerequisites

• ffmpeg>=2.1 with libx264 enabled
  • If you are using conda, you can install the correct version with conda remove ffmpeg and conda install ffmpeg x264 -c conda-forge
• ffprobe (usually comes with ffmpeg)

Quickstart

Save/load RGB frames:

import numpy as np
from videoio import videosave, videoread

frames = np.random.random((20, 200, 400, 3))  # [framesNr, height, width, RGB]
# Save to video
videosave("out.mp4", frames)
# Load from video
frames = videoread("out.mp4")

Read frames sequentially:

from videoio import VideoReader

for frame in VideoReader("in.mp4"):
    do_something_with(frame)

Write frames sequentially:

from videoio import VideoWriter

writer = VideoWriter("out.mp4", resolution=(400, 200))  # [width, height]
for i in range(100):
    frame = get_frame()
    writer.write(frame)
writer.close()

or

with VideoWriter("out.mp4", resolution=(400, 200)) as writer:
    for i in range(100):
        frame = get_frame()
        writer.write(frame)

Lossless write/read of uint16 3D arrays (useful for saving depth frames stored in mm, for example Kinect data):

import numpy as np
from videoio import uint16save, uint16read

# Generate 20 random depth frames
depth_frames = (np.random.random((20, 200, 400)) * 65535).astype(np.uint16)
# Save
uint16save("out_depth.mp4", depth_frames)
# Load
depth_frames = uint16read("out_depth.mp4")

Save RGB frames in lossless mode with different compression preset and different FPS:

videosave("out.mp4", frames, lossless=True, preset="veryfast", fps=10.5)

Read RGB frames and scale them to target resolution simultaneously:

frames = videoread("in.mp4", output_resolution=(100, 250))

Read video/uint16-array starting from certain frame:

(Works if the input video was created by videoio, other cases are not guaranteed)

frames = videoread("in.mp4", start_frame=100)

for frame in VideoReader("in.mp4", start_frame=100):
    do_something_with(frame)

Installation

From pip:

pip install videoio

From source:

git clone https://github.com/vguzov/videoio.git
python setup.py install