Adaptive Semantic Communication for Gaze-guided Stereo Media Transmission

Stereo images, which comprise image pairs captured from slightly offset viewpoints to simulate binocular vision, have found extensive applications in domains such as virtual reality (VR), 3D reconstruction, and autonomous driving. Considering their integration in immersive systems like VR head-mounted displays, this work proposes a gaze-guided adaptive semantic communication for stereoscopic media transmission. By dynamically adjusting the transmission of visual features in response to varying channel conditions, the proposed method prioritizes the semantic content most relevant to the user’s gaze. This enables efficient rate adaptation and ensures high-quality reconstruction of regions of interest, thereby enhancing perceptual experience under constrained bandwidth scenarios.