首页|Reports Summarize Robotics and Automation Study Results from University of New S outh Wales (Fully Decoupling Trajectory and Scene Encoding for Lightweight Heatm ap-oriented Trajectory Prediction)
Reports Summarize Robotics and Automation Study Results from University of New S outh Wales (Fully Decoupling Trajectory and Scene Encoding for Lightweight Heatm ap-oriented Trajectory Prediction)
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News Daily News – Fresh data on Robotics - Robotics and Automation are presented in a new report. According to news reporting out of Syd ney, Australia, by NewsRx editors, research stated, “Recently, heatmap-oriented approaches have demonstrated their state-of-the-art performance in pedestrian tr ajectory prediction by exploiting scene information from input images before run ning the encoder. To align the image and trajectory information, existing method s centre the scene images to agents’ last observed locations or convert trajecto ry sequences into images.” Our news journalists obtained a quote from the research from the University of N ew South Wales, “Such alignment processes cause repetitive executions of the sce ne encoder for each pedestrian in an input image while there are often many pede strians in an image, thus leading to significant memory consumption. In this let ter, we address this problem by fully decoupling scene and trajectory feature ex tractions so that the scene information is only encoded once for an input image regardless of the number of pedestrians in the image. To do this, we directly ex tract temporal information from trajectories in a global pixel coordinate system . Then, we propose a transformer-based heatmap decoder to model the complex inte raction between high-level trajectory and image features via trajectory self-att ention, trajectory-to-image cross-attention and image-to-trajectory cross-attent ion layers. We also introduce scene counterfactual learning to alleviate the ove r-focusing on the trajectory features and knowledge transfer from Segment Anythi ng Model to simplify the training.”
SydneyAustraliaAustralia and New Zea landRobotics and AutomationRoboticsUniversity of New South Wales
NETL
NSTL
