An advanced recurrent neural network (RNN) implemented in bash, mainly useful to extract scores from Youtube music videos.
To run this advanced Artificial Intelligence, you will need to install the following:
- youtube-dl - https://github.com/ytdl-org/youtube-dl
- imagemagick - https://github.com/ImageMagick/ImageMagick"
- ffmpeg - https://github.com/FFmpeg/FFmpeg"
- findimagedupes - https://github.com/opennota/findimagedupes"
Installing sheet-extractor.sh is as simple as cloning this repo on your local machine.
./sheet-extractor.sh https://youtu.be/cUkHdt2im6M
This will:
- Download the video (using youtubedl)
- Extract a number of frames from the video (using ffmpeg)
- Discard similar frames (using findimagedupes)
- Merge the remaining frames in a single pdf (using imagemagick)
The resulting file will be the following: TODO
It may happen that some the resulting PDF has duplicate pages. In that case, I would advise you to modify the accuracy of findimagedupes like this:
./sheet-extractor.sh 90% https://youtu.be/cUkHdt2im6M
98% accuracy means that two frames are considered the same only when they are 98% equal. The correct accuracy largely depends on the video downloaded; the default is set to 95%.
This AI is trained to work best when the video is made of still images only. It may still work when there is a moving line indicating the current position in the score, but only if the score doesn't scroll too.
Do not use this to download copyrighted scores. Use this small script ethically.