Servicenow Canada Inc.

Element Ai Inc.

Negin Sokhandan Asl

Perouz Taslakian

A system identifying anomalies in an image of an object is first trained using first sets of images corresponding to first anomaly types for the object. A model of the object is formed in a latent space. A label for each anomalous image is used to calculate vectors containing means and standard deviations for each first anomaly types. The means and standard deviations are used to calculate a log-likelihood loss for each first anomaly type. The system is retrained using second sets of images corresponding to second anomaly types for the object. The vectors are supplemented using labels for each second anomaly types. A statistically sufficient sample of information in the means and standard deviations vectors is supplied to the latent space. A log-likelihood loss for each of the first and second anomaly types is calculated based on their respective mean and standard deviation.

Continuous training methods for systems identifying anomalies in an image of an object

Systems and methods for detecting and predicting text within images. An image is passed to a feature-extraction module. Each image typically contains at least one text object, and each text object contains at least one character. Based on the image, the feature-extraction module generates at least one feature map indicating text object(s) in the image. The feature map(s) is then passed to a decoder module. In son implementations, the decoder module applies a weighted mask to the feature map(s). Based on the feature map(s), the decoder module predicts a sequence of characters in the text object(s). In some embodiments, that prediction is based on previous known data. The decoder module is directed by a query that indicates at least one desired characteristic of the text object(s). An output module then refines the predicted content. At least one neural network may be used.

Automatically predicting text in images

A system identifies anomalies in an image of an object. An input image of the object containing zero or more anomalies is supplied to an image encoder. The image encoder generates an image model. The image model is applied to an image decoder that forms a substitute non-anomalous image of the object. Differences between the input image and the substitute non-anomalous image identify zero or more areas of the input image that contain the zero or more the anomalies. The system implements a flow-based model and has been trained using (a) a set of augmented anomaly-free images of the object applied at the image encoder and (b) a reconstruction loss calculated based on a norm of differences between each augmented anomaly-free image of the object and a corresponding output image from the image decoder.

Systems and computer-implemented methods for identifying anomalies in an object and training methods therefor

Systems and methods for monitoring product placement. The method comprises accessing a first image depicting a plurality of items arranged in accordance with a first layout, accessing a second image, the second image depicting at least some of the plurality of items arranged in accordance with a second layout. The method then proceeds to inputting, to a machine learning algorithm (MLA), a first density map and a second density map, the first density map having been generated from the first image and the second density map having been generated from the second image. An anomaly map is then outputted by the ML, the anomaly map comprising a first indication of an item class associated with an anomaly and a second indication of a position associated with the anomaly.

Growing community of inventors

Montreal, Canada

Negin Sokhandan Asl