DiffAttn: Diffusion-Based Drivers' Visual Attention Prediction with LLM-Enhanced Semantic Reasoning
Researchers have developed DiffAttn, a novel diffusion-based framework for predicting drivers' visual attention. This system integrates a Swin Transformer for scene feature extraction and a Feature Fusion Pyramid for enhanced denoising and context modeling. A key innovation is the incorporation of a large language model (LLM) layer to improve semantic reasoning and identify safety-critical cues. Experiments on multiple datasets show DiffAttn outperforms existing methods, offering potential for improved intelligent vehicle safety and driver understanding. AI
IMPACT This research could lead to more sophisticated driver-assistance systems by improving how vehicles understand and anticipate human visual focus.