OBJECT POSE ESTIMATION FROM AN RGB IMAGE BY A NEURAL NETWORK OUTPUTTING A PARAMETER MAP

摘要

A method for processing at least one image comprising inputting the image to at least one neural network (ANN1, ANN2), the at least one network being configured to deliver, for each pixel of a group of pixels belonging to an object of a given type visible on the image, an estimation of parameters of the object,the method further comprising processing the estimations of the parameters of the object using an instance segmentation mask identifying instances of objects having the given type.