write_video

Usage

Writes video data in tensor format to a file.

Interface

torchvision.io.write_video(filename, video_array, fps, video_codec='libx264', options=None, audio_array=None, audio_fps=None, audio_codec='aac', audio_options=None)

Parameters

Parameter	Description	Value Range	Input/Output
filename	Path to the output video file.	Character string	Input
video_array	Video frame array, with a shape of (T, C, H, W). T indicates the number of frames, C indicates the number of channels (usually 3, standing for RGB), and H and W indicate the height and width, respectively.	torch.Tensor	Input
fps	Video frame rate (frames per second).	(0, INT_MAX]	Input
video_codec	Video encoder. Multiple encoders are available, such as libx264 (default), libx265, and mpeg4.	Character string	Input
options	Extra options of the video encoder, provided in dictionary format. For example, {'crf': '23'} sets the constant rate factor.	Dictionary	Input
audio_array	Audio frame array, with a shape of (T, N). T indicates the number of audio frames, and N indicates the number of audio channels.	torch.Tensor	Input
audio_fps	Audio sample rate (frames per second).	(0, INT_MAX]	Input
audio_codec	Audio encoder. Multiple encoders are available, such as aac (default) and mp3.	Character string	Input
audio_options	Extra options of the audio encoder, provided in dictionary format. For example, {'b:a': '128k'} sets the audio bit rate.	Dictionary	Input

Example

import torch
import torchvision.io as io

# Generate a simple video frame array.
video_array = torch.rand((100, 3, 1080, 1920))  # 100 frames, 1080p, RGB

# Generate a simple audio frame array.
audio_array = torch.rand((100000, 2))  # 100,000 samples, dual-channel

# Define the path to the output file.
filename = 'output_video.mp4'

# Write a video file.
io.write_video(
    filename=filename,
    video_array=video_array,
    fps=30,
    video_codec='libx264',
    options={'crf': '23'},
    audio_array=audio_array,
    audio_fps=44100,
    audio_codec='aac',
    audio_options={'b:a': '128k'}
)

Parent topic: Interface Definition