MObyGaze: a film dataset of multimodal objectification densely annotated by experts

TL;DR

This paper introduces MObyGaze, a densely annotated multimodal film dataset for analyzing objectification, and explores AI methods to characterize and quantify complex gender representation patterns in movies.

Contribution

It presents a new multimodal dataset with expert annotations on objectification in films and investigates learning approaches for this complex, multi-label task.

Findings

Feasibility of modeling objectification using multimodal data

Benchmark results for vision, text, and audio models on the dataset

Effective methods for learning from diverse, expert-annotated labels

Abstract

Characterizing and quantifying gender representation disparities in audiovisual storytelling contents is necessary to grasp how stereotypes may perpetuate on screen. In this article, we consider the high-level construct of objectification and introduce a new AI task to the ML community: characterize and quantify complex multimodal (visual, speech, audio) temporal patterns producing objectification in films. Building on film studies and psychology, we define the construct of objectification in a structured thesaurus involving 5 sub-constructs manifesting through 11 concepts spanning 3 modalities. We introduce the Multimodal Objectifying Gaze (MObyGaze) dataset, made of 20 movies annotated densely by experts for objectification levels and concepts over freely delimited segments: it amounts to 6072 segments over 43 hours of video with fine-grained localization and categorization. We formulate different learning tasks, propose and investigate best ways to learn from the diversity of labels among a low number of annotators, and benchmark recent vision, text and audio models, showing the feasibility of the task. We make our code and our dataset available to the community and described in the Croissant format: https://anonymous.4open.science/r/MObyGaze-F600/.