Beyond Real versus Fake Towards Intent-Aware Video Analysis

TL;DR

This paper introduces IntentHQ, a comprehensive benchmark for analyzing the underlying intent behind videos, shifting focus from authenticity detection to understanding motivations using multi-modal models.

Contribution

The paper presents IntentHQ, a new annotated video dataset with fine-grained intent categories and a multi-modal approach for intent recognition, advancing human-centered video analysis.

Findings

IntentHQ contains 5168 videos with 23 intent categories.

Multi-modal models effectively recognize diverse video intents.

The approach improves understanding of video motivations beyond fake detection.

Abstract

The rapid advancement of generative models has led to increasingly realistic deepfake videos, posing significant societal and security risks. While existing detection methods focus on distinguishing real from fake videos, such approaches fail to address a fundamental question: What is the intent behind a manipulated video? Towards addressing this question, we introduce IntentHQ: a new benchmark for human-centered intent analysis, shifting the paradigm from authenticity verification to contextual understanding of videos. IntentHQ consists of 5168 videos that have been meticulously collected and annotated with 23 fine-grained intent-categories, including "Financial fraud", "Indirect marketing", "Political propaganda", as well as "Fear mongering". We perform intent recognition with supervised and self-supervised multi-modality models that integrate spatio-temporal video features, audio processing, and text analysis to infer underlying motivations and goals behind videos. Our proposed model is streamlined to differentiate between a wide range of intent-categories.