This repository provides a moudle to download the original videos in Youtube-8m dataset
Since the official youtube-8m dataset
website contains only videos and frame level features in the format of tensorflow protocal buffers. Hence, in this repository I write a tool to download the orignal vedios.
Dependencies for downloading youtube video ids for categories
pip install requests pytube progressbar
or
conda install requests progressbar
conda install -c everwho pytube
- Open
categories.txt
. - Select the categories and paste them into
downloadlist.txt
. Note, there is only one category for each line and the first letter of each category is Capitalized. - Save
downloadlist.txt
.
python downloader.py
The IDs of each category are saved at the folder 'ID'. The file of ID are named as the categories.
The Vedios of each category are saved at the folder 'vedios\YOUR CATEGORY NAME'. By default a video is downloaded in the best possible resolution.
Please see pytube
If you have any questions or suggestions about the code, feel free to create an issue.