Method And System For Training A Convolution Neural Network (Cnn) And Inferencing Using Trained Cnn

Abstract: A method and a system for training a convolution neural network (CNN) for performing one or more computer vision tasks on an image is disclosed. The method comprises selecting one or more computer vision tasks to be performed on an image `and performing a plurality of convolution operations on the image using a plurality of convolution layers to generate a feature map for the image. Said method further discloses dropping one or more features from the feature map and performing the one or more computer vision task on the image based on remaining features. Said method then compares result of each of the one or more computer vision tasks with respective ground truth result to calculate individual losses and updates weights based on a weighted sum loss calculated from the individual losses, which is used as feedback to the plurality of convolutional layers for training the CNN.

Patent Information

Applicants

Inventors

Specification

Documents

Application Documents

Search Strategy

ERegister / Renewals Inforce