Learning to Cut by Watching Movies
October 19, 2021 ยท View on GitHub
The data from this project was downloaded from youtube.
Original videos
The original videos were collected from the YouTube Channel: Movieclips.
To access the original videos check their youtube ID under the column video_id in the files: subset_moviescenes_shotcuts_{train,val}.csv.
Then, go to youtube an replace the id in this URL: https://www.youtube.com/watch?v={video_id}
Feature Extraction
The features used for the paper are shared in the main README.md:
| Resource | Video | Audio | Destination Folder |
|---|---|---|---|
| Features | link | link | ./data/ |
However, if you wanna extract features yourself. You can follow the repos below:
| Resource | Features | Notes |
|---|---|---|
| Video | ResNexT-101 Kinetics | Snippet size: 16 Frames at 24FPS with 8-frame overlapping windows |
| Audio | ResNet-18 VGGSound | Parameters used can be found in the repo |