kinetics-download

Simple tool to download videos using proxy from kinetics dataset.

Also have functionality to trim downloaded videos to the length of action. To support trimming, ffmpeg should be installed and added to environment variable PATH.

Usage

Install requirements first

pip install -r requirements.txt

Install youtube-dl and ffmpeg according to your system and added them to environment variable PATH.

You should download and unzip csv files with links to videos. You can download such files here. For example here is link to kinetics-600 training.zip

Downloading

failure_list.txt stores unavaliable links.

cd kinetics-download
python download.py failure_list.txt /path/to/kinetic_train.csv /path/to/videos/

Downloading and trimming

cd kinetics-download
python download.py failure_list.txt /path/to/kinetic_train.csv /path/to/videos/ --trim

Parallel processing

One can run downloading and trimming in several processes. To do so, just add flag --num-jobs with number of jobs running in parallel. For example:

cd kinetics-download
python download.py failure_list.txt /path/to/kinetic_train.csv /path/to/videos/ --trim --num-jobs 10

Trimming with your Nvidia device

You can execute ffmpeg trimming with cuda accelerating:

cd kinetics-download
python download.py failure_list.txt /path/to/kinetic_train.csv /path/to/videos/ --trim --num-jobs 10 --NV

With proxy

We add the specified HTTP/HTTPS/SOCKS proxy support for users cannot visit YouTube by direct connection. To enable SOCKS proxy, specify a proper scheme. For example socks5://127.0.0.1:1080/. Pass in an empty string (--proxy "") for direct connection'). For example:

cd kinetics-download
python download.py failure_list.txt /path/to/kinetic_train.csv /path/to/videos/ --trim --num-jobs 10 --proxy https://127.0.0.1:8087

Reference

This project is mostly derived from https://github.com/piaxar/kinetics-downloader