MSN: Multi-Style Network for Trajectory Prediction

It is essential to predict future trajectories of various agents in complex scenes. Whether it is internal personality factors of agents, interactive behavior of the neighborhood, or the influence of surroundings, it will have an impact on their future plannings. It means that even for the same physical type of agents, there are huge differences in their behavior styles. We concentrate on the problem of modeling agents' multi-style characteristics when predicting their trajectories. We propose the Multi-Style Network (MSN) to focus on this problem by dividing agents' behaviors into several hidden behavior categories adaptively, and then train each category's prediction network jointly, thus giving agents multiple styles of predictions simultaneously. Experiments show that MSN outperforms current state-of-the-art methods with 10\% - 20\% performance improvement on two widely used datasets, and presents better multi-style characteristics in predictions.