This repository contains:
- An implementation of the 2015 paper in the ACM Transactions on Graphics by Jain et al.
Their project page can be seen at http://www.cise.ufl.edu/~ejain/VideoRe-editing.html
- A work in progress to improve the system (l1-optimization directory), which can model 2-D movement.
To run the implementation:-
Run minimization.m with the appropriate csv file in the csv_files directory
Each entry in the csv file corresponds to eye tracking data observed for a frame in the original video.
The output is a plot of the final cropping window path along with the output frames.
In croppath.m change the directory path to folder which contains the input frames.
Run l1-main.m in l1-optimization directory to test the second model.
The implementation has been tested on test videos of different scene types.
Test videos:-
- Herbie Rides Again (waterfront scene)
- Herbie Rides Again (chauffeur scene)
- Analyze This (Hotel room conversation)
- Analyze This (Hotel Shooting)
- Fast Five (Stealing the vault)
- Tranformers 3 (Highway chase)
- Transformers 3 (Slow motion scene)
- Batman vs Superman trailer
- Brazil vs Germany 3rd goal
- Matrix clip-1
- Matrix clip-2
- Harry Potter scene
- Scene from a French theater drama