Results on different parts of algorithm. Each line represents different value of minimum interval value (--min-scenecut command line parameter).
It can be seen that performance of the algorithm on difficult youtube videos is worse than on documental films from BBC.
F score can be calculated using formula below:
where Precision and Recall are:
tp is number of correctly detected scene changes
tn is number of correctly detected frames without scene changes
fp is number of false alarms
fn is number of missed scene changes
The average speed for the algorithm:
Part | FPS |
---|---|
4K videos | 2.5 |
BBC Dataset | 238 |
Other videos | 30 |
TO DO:
Dependency of F score and speed on resolution
Top comments (0)