Techniques for quantifying accuracy of a prediction model that has been trained on a data set parameterized by multiple features are provided. The model performs in accordance with a theoretical performance manifold over an intractable input space in connection with the features. A determination is made as to which of the features are strongly correlated with performance of the model. Based on the features determined to be strongly correlated with performance of the model, parameterized sub-models are created such that, in aggregate, they approximate the intractable input space. Prototype exemplars are generated for each of the created sub-models, with the prototype exemplars for each created sub-model being objects to which the model can be applied to result in a match with the respective sub-model. The accuracy of the model is quantified using the generated prototype exemplars. A recommendation engine is provided for when there are particular areas of interest.

A method for detecting an object based on a video includes: obtaining a plurality of image frames of a video to be detected; obtaining initial feature maps by extracting features of the plurality of image frames; for each two adjacent image frames of the plurality of image frames, obtaining a target feature map of a latter image frame of the two adjacent image frames by performing feature fusing on the sub-feature maps of the first target dimensions included in the initial feature map of a former image frame of the two adjacent image frames and the sub-feature maps of the second target dimensions included in the initial feature map of the latter image frame; and performing object detection on the respective target feature map of each image frame.

A method for recognizing a text, an electronic device and a storage medium. An implementation of the method comprises: obtaining a multi-dimensional first feature map of a to-be-recognized image; performing, based on feature values in the first feature map, feature enhancement processing on each feature value in the first feature map; and performing a text recognition on the to-be-recognized image based on the first feature map after the enhancement processing.

A device includes image generation circuitry and convolutional-neural-network circuitry. The image generation circuitry, in operation, generates a digital image representation of a wafer defect map (WDM). The convolutional-neural-network circuitry, in operation, generates a defect classification associated with the WDM based on: the digital image representation of the WDM and a data-driven model associating WDM images with classes of a defined set of classes of wafer defects and generated using a training data set augmented based on defect pattern orientation types associated with training images.

According to an embodiment of the disclosure, an electronic device may include: a display, a memory, and a processor operatively connected to the display and the memory. According to an embodiment, the memory may store instructions that, when executed, cause the processor to: obtain a first image of a first shape, obtain linear information indicating a morphological characteristic of an object in the first image of the first shape, determine a conversion method for converting the first image of the first shape into an image of a second shape based on the obtained linear information, convert the first image of the first shape into a second image of the second shape based on the determined conversion method, and control the display to display the converted second image of the second shape on the display.

A system and method to compress application control data, such as weights for a layer of a convolutional neural network, is disclosed. A multi-core system for executing at least one layer of the convolutional neural network includes a storage device storing a compressed weight matrix of a set of weights of the at least one layer of the convolutional network and a decompression matrix. The compressed weight matrix is formed by matrix factorization and quantization of a floating point value of each weight to a floating point format. A decompression module is operable to obtain an approximation of the weight values by decompressing the compressed weight matrix through the decompression matrix. A plurality of cores executes the at least one layer of the convolutional neural network with the approximation of weight values to produce an inference output.

Provided herein are systems and methods for automated identification of volumes of interest in volumetric brain images using artificial intelligence (AI) enhanced imaging to diagnose and treat acute stroke. The methods can include receiving image data of a brain having header data and voxel values that represent an interruption in blood supply of the brain when imaged, extracting the header data from the image data, populating an array of cells with the voxel values, applying a segmenting analysis to the array to generate a segmented array, applying a morphological neighborhood analysis to the segmented array to generate a features relationship array, where the features relationship array includes features of interest in the brain indicative of stroke, identifying three-dimensional (3D) connected volumes of interest in the features relationship array, and generating output, for display at a user device, indicating the identified 3D volumes of interest.

This application discloses an image content removal method, and relates to the field of computer vision. The method includes: enabling a camera application; displaying a photographing preview interface of the camera application; obtaining a first preview picture and a first reference frame picture that are captured by a camera; determining a first object in the first preview picture as a to-be-removed object; and determining to-be-filled content in the first preview picture based on the first reference frame picture, where the to-be-filled content is image content that is of a second object and that is shielded by the first object in the first preview picture. The terminal generates a first restored picture based on the to-be-filled content and the first preview picture. In this way, image content that a user does not want in a picture or a video shot by the user can be removed.

In a learning device, a feature extraction means extracts image features from an input image. A class discrimination means discriminate a class of the input image based on the image features, and generates a class discriminative result. A class discriminative loss calculation means calculates a class discriminative loss based on the class discriminative result. A normal/abnormal discrimination means discriminates whether the class is a normal class or an abnormal class, based on the image features, and generates a normal/abnormal discriminative result. The AUC loss calculation means calculates an AUC loss based on the normal/abnormal result. A first learning means updates parameters of the feature extraction means, a class discrimination means, and the normal/abnormal discrimination means, based on the class discriminative loss and the AUC loss.

The training apparatus (2000) performs a first phase training and a second phase training of a discriminator (10). The discriminator (10) acquires a ground-view image and an aerial-view image, and determines whether the acquired ground-view image matches the acquired aerial-view image. The first phase training is performed using a ground-view image and a first level negative example of aerial-view image. The first level negative example of aerial-view image includes scenery of a different type from scenery in the ground-view image. The second phase training is performed using the ground-view image and a second level negative example of aerial-view image. The second level negative example of aerial-view image includes scenery of a same type as scenery in the ground-view image.