14 Apr 2017 | Namhoon Lee, Wongun Choi, Paul Vernaza, Christopher B. Choy, Philip H. S. Torr, Manmohan Chandraker
The paper introduces DESIRE, a Deep Stochastic IOC RNN Encoder-decoder framework for predicting the future locations of multiple interacting agents in dynamic scenes. DESIRE addresses the multi-modal nature of future predictions, strategic decision-making based on potential outcomes, and reasoning from past motion history, scene context, and agent interactions. The model uses a conditional variational auto-encoder (CVAE) to generate diverse future prediction samples, an RNN scoring-regression module to rank and refine these samples based on accumulated future rewards, and a Scene Context Fusion (SCF) module to integrate past motion, semantic scene context, and agent interactions. The model is evaluated on the KITTI and Stanford Drone datasets, showing significant improvements over baseline methods in prediction accuracy. Key contributions include the ability to handle complex, dynamic scenes, generate diverse hypotheses, and make strategic decisions while considering long-term future rewards.The paper introduces DESIRE, a Deep Stochastic IOC RNN Encoder-decoder framework for predicting the future locations of multiple interacting agents in dynamic scenes. DESIRE addresses the multi-modal nature of future predictions, strategic decision-making based on potential outcomes, and reasoning from past motion history, scene context, and agent interactions. The model uses a conditional variational auto-encoder (CVAE) to generate diverse future prediction samples, an RNN scoring-regression module to rank and refine these samples based on accumulated future rewards, and a Scene Context Fusion (SCF) module to integrate past motion, semantic scene context, and agent interactions. The model is evaluated on the KITTI and Stanford Drone datasets, showing significant improvements over baseline methods in prediction accuracy. Key contributions include the ability to handle complex, dynamic scenes, generate diverse hypotheses, and make strategic decisions while considering long-term future rewards.