Detection of Animated Scenes Among Movie Trailers

EasyChair Preprint 7815

Abstract

This paper presents a method for the detection of animated scenes in movie trailers. Regardless of the studio, artists, and the unique features of diverse animation creation techniques, machine learning tools can provide a concise detection methods of animated scenes in movie components. A dataset is prepared by selecting scenes from trailers us- ing shot boundary analysis and removal of scenes containing non-movie contents; e.g. credentials. The dataset is composed of over 1400 movie scenes from 230 trailers of various genre. A Convolutional Neural Net- work (CNN) architecture followed by Gated Recurrent Unit (GRU) is built by using transfer learning of EfficientNet.

Keyphrases: Movie Trailer Labeling, Scene Understanding, computer vision, deep learning

@booklet{EasyChair:7815,
  author    = {Irmak Türköz and Halil Altay Güvenir},
  title     = {Detection of Animated Scenes Among Movie Trailers},
  howpublished = {EasyChair Preprint 7815},
  year      = {EasyChair, 2022}}