Machine detection of film edits

A film can be fundamentally broken into innumerous shots, placed after one another. These shots are divided by cuts. Film cuts can be broadly divided into two categories - Sharp/Hard cuts, Gradual cuts. This work is about the detection and classification of these film cuts. While there are many algorithms for cut detection, few of them yield a good performance efficiency ratio. In this work, a synthetic dataset has also been made from scratch using appropriate algorithms. A method has been used in which the hard cuts and gradual cuts are filtered and processed in different modules using different deep learning techniques. This approach should yield a descent performance and less computational complexity. The proposed model is capable of detecting multiple cuts in a video and return timestamps of the cut occurrence in the video. I have also proposed on making a flask webapp/ pyQT GUI to integrate this model at the backend of the webapp or GUI. The webapp can be added to RedHenLab’s rapid annotator.