Layar B.v.

Tomtom Global Content B.v.

Klaus Michael Hofmann

Dirk Groten

Mohsen Ghafoorian

Anatoliy Samara

Vladimir Nedovic

Raimo Juhani Van Der Klein

Erik Stammes

The invention relates to a method of training a model for use in semantic image segmentation. The model includes a first classifier neural network and a second classifier neural network. The method includes training the first network by inputting a first training image containing a target object to the first network, using the first network to identify and erase pixels of the first training image that are discriminative for the target object, inputting the first training image to the second network, using the second network to determine a likelihood of the first modified training image containing the target object and updating weights of the first network using a first loss function that is a monotonically-increasing function of the determined likelihood.

Neural network models for semantic image segmentation

Method and systems for generating reference features sets for slices of a reference image. The reference features sets generated from slices enables better object recognition and/or tracking when a camera image only shows a portion of the reference image. Metadata is used to link the reference features set of the original image and of the slices together as belonging to the same object, providing hierarchical relationship information and/or spatial relationship information. An image processing function may be dynamically configured on the basis of whether an object has been successfully detected and the metadata associated with the object.

Reference image slicing

A method for improving the placement of virtual content in augmented reality is disclosed. A reference image of the object to be augmented is analyzed for feature density information. The placement of content is based on the at least one location adjacent to where features are concentrated. Accordingly, the chance of the content interfering with an important part of the object is lowered, and object recognition and/or tracking ability is maintained because there are sufficient features nearby the content to enable object recognition and/or tracking.

Determining space to display content in augmented reality

Disclosed herein is an improved method for providing content associated with barcodes in augmented reality in addition or in combination with providing content associated with target objects in augmented reality. The improved method advantageously provides a augmented reality client that a user may use to view the respective content associated with barcodes and target objects while in camera view to improve usability. Advantageously, the user is not unexpectedly taken out of camera view to view the content associated with the barcode and the user experience provided is consistent between barcodes and target objects. Furthermore, the improved method integrates barcodes and a visualization of the barcode within augmented reality, without disrupting the real-time augmented reality experience.

Barcode visualization in augmented reality

Methods and systems for determining augmentability information associated with an image frame captured by a digital imaging part of a user device. The determined augmentability score may then be used in the generation of feedback to the user. For example, a graphical user interface may be generated and rendered having a substantially continuous visual output corresponding to the augmentability information.

Growing community of inventors

Amsterdam, Netherlands

Klaus Michael Hofmann