METHOD AND DEVICE FOR TRAINING A NEURAL NETWORK TO SPECIFY LANDMARKS ON 2D AND 3D IMAGES

摘要

The present invention relates generally to a method, device and computer program product for training neural networks configured to generate specifications of landmarks related to two-dimensional or three-dimensional image data. The specification of landmarks is defined by three vectors with position, precision and presence values forming landmark specification vectors. The neural network extracts feature vector from input image data using convolutional neural network layers, decodes the feature vector, and outputs landmark specification vectors. The present invention further relates to a system of devices for specifying landmarks on images generated by neural networks running on the devices.