Skip to content

vggsound

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

vggsound

README.md

Downloading the VGGSound Dataset

This folder contains the scripts for downloading the VGGSound dataset. The CSV file is downloaded from the original repository.

Prerequisites

Shuffle and split the CSV file as follows. Put vggsound.csv into your data directory, e.g. data/vggsound.

shuf data/vggsound/vggsound.csv > data/vggsound/vggsound-shuf.csv
split -l 10000 -d --additional-suffix=.csv data/vggsound/vggsound-shuf.csv data/vggsound/vggsound-shuf-

Install packeages

pip install youtube_dl tqdm pafy

Download the dataset

Run the following script over all the CSV files.

python download_ffmpeg.py -e -s -i data/vggsound/vggsound-shuf-00.csv -o data/vggsound/video/00/

Extract audio from videos

python extract_audio.py -i data/vggsound/video -o data/vggsound/audio -s -e

Extract image frames from videos

python extract_frames.py -i data/vggsound/video -o data/vggsound/frames -s -e

Resize and crop images

python preprocess.py -i data/vggsound/frames -o data/vggsound/preprocessed -s -e