Skip to content

Latest commit

 

History

History
73 lines (48 loc) · 2.05 KB

README.md

File metadata and controls

73 lines (48 loc) · 2.05 KB
Raw

Youtube video scene search

Open In Colab

This repository was created to find scenes that you waant auotomatically in youtube video.

I provide automation tool to find specific scenes with your text.

Environment(conda)

Deep learning models are used in this tool.

Models : CLIP, BLIP-2

So I recommend you to install the following command, for setting an environment.(Also You can run in Colab)

Usage

  1. Clone repo from here https://github.com/gyuilLim/youtube_scene_search
  2. Change the directory to where you have cloned repo.
cd youtube_scene_search
  1. Create conda virtual environment.
conda create -n youtube_scene_search python=3.8
conda install --file requirements.txt
  1. Install the command.

I give you CLI for making it easy to execute.

CLI requires 5 aurguments.

  • youtube video(str) : video that you want to find sepecific scenes.
  • text(str) : text that you want to search with in youtube video.
  • model(str) : CLIP or BLIP.
  • kfe(bool) : kfe means 'key frame extraction'. you can choose whether to use it or not.
  • download(bool) : whether to download youtube video.

So, you can install as CLI like this :

$ python text_search_main.py \
--url 'https://www.youtube.com/watch?v=82C19hXaloc' \
--text "A man standing while holding a book" \
--model "clip" \
--kfe False \
--download True

Return value is the youtube link where the scene is played.

https://www.youtube.com/watch?v=82C19hXaloc&t=4

Result

image

Scene that is matched with input text "A man standing while holding a book"

Recommended specifications

CLIP BLIP-2
GPU memory 4.5GB 12GB

You can run this tool in the Colab T4 (15GB) environment.