Novel Video Prediction for Large-scale Scene using Optical Flow

• DOI: 10.48550/arxiv.1805.12243
• Publication Date: 2018-01-01
• Authors: Henglai Wei; Xiaochuan Yin; Penghong lin
• Keywords: Optical Flow

Making predictions of future frames is a critical challenge in autonomous driving research. Most the existing methods for video prediction attempt to generate simple and fixed scenes. In this paper, we propose novel effective optical flow conditioned method task with an application complex urban contrast previous work, model only requires sequences training testing. Our uses rich spatial-temporal features sequences. The takes advantage motion information extracting from maps between neighbor images as well images. Empirical evaluations on KITTI dataset Cityscapes demonstrate effectiveness our method.