A team of researchers from the University of Virginia recently performed a large-scale analysis aimed at identifying the characteristics of movie trailers that best predict a movie’s genre and estimated budget. In their study, described in a pre-published article on arXiv, the researchers specifically compared the effectiveness of visual, audio, text, and metadata-based features.
“Understanding video is the next frontier after understanding images,” said Vicente Ordonez, one of the researchers who conducted the study. TechXplore. “However, a lot of work on understanding video has so far focused on short clips with a human performing a single action. We wanted something longer, but there’s also the issue of the power of The video trailers seemed like an intermediate compromise, as they display a myriad of things, from scary to funny. “
Movie trailers are short and can easily be combined with movie descriptions. Ordonez and his colleagues realized that these characteristics make them ideal for studying the parallels between video and language.
In addition, recent studies have introduced several promising tools for analyzing images associated with textual descriptions. The researchers were curious to evaluate some of these techniques on video recognition tasks.
Initially, when they tried to apply well-established methods of analyzing short video clips to movie trailers, the results were disappointing. So they decided to conduct a thorough investigation to identify the most effective features for analyzing movie trailers.
“We found that by combining all modalities (i.e. video, text, audio and metadata), we were able to gather valuable information about the expected correlations between specific genres and a particular modality, for example , that visual features are more valuable when predicting an animated or non-animated movie, ”Paola Cascante-Bonilla, another researcher involved in the study, told TechXplore. “Additionally, we have found that including audio in our experiments significantly improves gender prediction performance compared to using only video, text, and metadata.”
The researchers observed that while analyzing movie posters gave unsatisfactory results, focusing on all of the movie features presented in a trailer (i.e. video, text, audio, and metadata) led to significant improvements. These results are particularly noteworthy, as they could help develop more efficient tools for analyzing films and serve as a basis for future research studies.
Interestingly, by focusing on the video, text, and audio data extracted from the trailers, Ordonez, Cascante-Bonilla, and their colleagues were able to estimate a movie’s genre with a precision comparable to that obtained by analyzing the movie’s metadata. (i.e. information about its actors, director, etc.). The techniques used by the researchers in their study, which combine different functionalities / modalities, could therefore be used to analyze a wider range of films.
In their study, the team also introduced a new dataset for training and evaluation of film analysis tools. This dataset, called Moviescope, includes 5,000 movies, along with their trailers, movie posters, movie storylines, and associated metadata.
“Our results suggest that a simple textual summary of a film is not sufficient to differentiate an animated film from a film of another genre,” said Siva Sivaraman, another researcher involved in the study. who now works at Microsoft. “You have to ‘see’ the trailer to be able to decide whether a particular movie is animated or not. The modal attention technique we used allows us to identify and analyze the characteristics that the model pays more attention to when predicting a particular gender. As we predicted, the model learns to weigh the visual characteristic against other characteristics while making predictions for the genre of animation. “
The results gathered by this team of researchers could have important implications for both film analysis and cinema advertising. In the future, other research groups could use these observations to develop more effective tools for predicting specific aspects of movies. Additionally, the techniques used by Ordonez and his colleagues could educate the advertising industry on how to create more impactful trailers.
“We now plan to use movie storylines and posters to analyze how movies are advertised and make recommendations to maximize the effectiveness of movie advertising from a consumer and distributor perspective,” Ordonez said. .
Moviescope: Large-scale analysis of movies using multiple modalities. arXiv: 1908.03180 [cs.CV]. arxiv.org/abs/1908.03180
© 2019 Science X Network
Quote: Survey of the best features for predicting a movie’s genre and estimated budget (August 30, 2019, retrieved October 20, 2021 from https://techxplore.com/news/2019-08-features-movie-genre.html
This document is subject to copyright. Other than fair use for private study or research purposes, no part may be reproduced without written permission. The content is provided for information only.