Nonlinear optimization for schooling deep learning purpose cloud analysis

Deep learning has improved the state-of the-art efficiency in wide use of computer vision-such as image identification ,object tracing segmentation and point cloud analysis.in this paper to comprehend Deep learning and improve performance and reason .we check optimization algorithm for education deep models and explore deep learning based point cloud analysis.in this paper we offer a approximation algorithm BPGrad.to optimize deep model via limb and recession .we offer a new approach to jointly know the ۳d subject model and guess the ۶D states of multifold example to the same object in a only end-to-end- deep neural network frame , with usage to profundity-based .sample segmentation.to the give the set of example point clouds with occlusions the network spontaneously take the block points in a certain camera view .wide experiment measure our style on several object models and changing number of instances in ۳d point clouds measure with popular baseline for sample segmentation ,our model not only show contesting efficiency ,but also read a ۳d object model that is show as a ۳d point cloud.

کلیدواژه ها:

Deep learning ، cloud analysis ، optimization ، BPGrad