Affecting Audience Valence and Arousal in 360 Immersive Environments: How Powerful Neural Style Transfer Is?

Immersive experiences in Virtual Reality (VR) platforms often require customized content that can be adapted to in-game situations and specific player actions that lead to individualized effects. Controlling the positive-negative emotional valence and the level of arousal of the immersive content allows VR systems to provide situation-specific and action-specific affective influence on players, giving them an experience that is tailor-made for the narrative and interaction espoused by the system. To generate different emotional influences for the same content, we created a system that uses Neural Style Transfer (NST) along with a set of style images with known affective ratings to procedurally generate different versions of the same 360 environments in VR with differing affective influences for players. To explore how the NST-generated VR affects participants’ affective perception, we conducted two user studies (N=30 and N=28). Users experienced four separate VR environments with different affective ratings. After each experience, we performed a survey to evaluate their affection, including Emotional Matching Tasks and interviews. Findings suggested that users are more likely to be aware of arousal differences than valence differences, which are mainly perceived by the degree of contrast between color and content of the environment. The stylized features gained from NST that affect the perception of valence are the color tone, the clarity of the texture, and the familiarity of the content for the user. Our work contributes novel insight into how users respond to generated VR environments and provides a machine-learning-based strategy for constructing an immersive environment to influence the affective experience of users, without altering any content and the game mechanism.