Abstract: Methods and systems for inspecting the integrity of multiple nested tubulars are included herein. A method for inspecting the integrity of multiple nested tubulars can comprise conveying an electromagnetic pipe inspection tool inside the innermost tubular of the multiple nested tubulars; taking measurements of the multiple nested tubulars with the electromagnetic pipe inspection tool; arranging the measurements into a response image representative of the tool response to the tubular integrity properties of the multiple nested tubulars; and feeding the response image to a pre-trained deep neural network (DNN) to produce a processed image, wherein the DNN comprises at least one convolutional layer, and wherein the processed image comprises a representation of the tubular integrity property of each individual tubular of the multiple nested tubulars.

Abstract: The present disclosure provides a determination method, an elimination method and an apparatus for an electron microscope aberration. The determination method comprises: training a neural network for image recognition using a plurality of electron microscope simulation images to obtain an electron microscope image recognition model; recognizing an electron microscope image of an experimental sample using the electron microscope image recognition model to obtain the electron microscope simulation image corresponding to the electron microscope image of the experimental sample; and obtaining the corresponding set aberration as an imaging aberration of the electron microscope image of the experimental sample according to the electron microscope simulation image corresponding to the electron microscope image of the experimental sample.

Abstract: A system for recording, storing and processing diagnostic information, including: a computer implementing a computer-readable media including digital data and ground truth; a registry constructed and arranged to store and associate transactions or accesses on the data; and a machine learning system that considers each learning step modification a microtransaction for the data used in that step and which is recorded in the transaction registry. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the methods.

Abstract: A method for generating one or more emoticons for one or more users with respect to one or more fictional characters is provided.

Abstract: An operating method of a computing apparatus is provided. The operating method of the computing apparatus includes obtaining a reference image; obtaining a distorted image generated from a reference image; obtaining an objective quality assessment score of a distorted image that is indicative of a quality of a distorted image as assessed by an algorithm, by using a reference image and a distorted image; obtaining a subjective quality assessment score corresponding to a objective quality assessment score; and training a neural network, by using a distorted image and a subjective quality assessment score as a training data set.

Abstract: Methods and systems for automated video segmentation are disclosed. A sequence of video frames having video segments of contextually-related sub-sequences may be received. Each frame may be labeled according to segment and segment class. A video graph may be constructed in which each node corresponds to a different frame, and each edge connects a different pair of nodes, and is associated with a time between video frames and a similarity metric of the connected frames. An artificial neural network (ANN) may be trained to predict both labels for the nodes and clusters of the nodes corresponding to predicted membership among the segments, using the video graph as input to the ANN, and ground-truth clusters of ground-truth labeled nodes. The ANN may be further trained to predict segment classes of the predicted clusters, using the segment classes as ground truths. The trained ANN may be configured for application runtime video sequences.

Abstract: A method performed by a computer is disclosed. The method comprises receiving color data for input pixels of an input image and an input set of features used to render the input image of a three-dimensional animation environment, wherein the input pixels are of a first resolution. The computer may then load into memory a generator of a generative adversarial network including a neural network used to scale the input image, the neural network trained using training data comprising color data of training input images and training output images and a training set of the features used to render the training input images. After the generator is loaded into memory, the computer may generate an output image having a second resolution that is different than the first resolution by passing the color data and the input set of features through the generator.

Abstract: This application provides a model inference method and apparatus based on a graphics rendering pipeline, and a storage medium for model inference based on a graphics rendering pipeline. The method includes: obtaining an instruction stream in a render thread; extracting and saving texture data information from the instruction stream, where the texture data information includes texture data; and inputting the texture data information to a graphics processing unit (GPU) rendering pipeline, where the GPU rendering pipeline is used to perform GPU model-based inference on the texture data based on a GPU model to obtain an inference result of the texture data, and the GPU model is a model running in a GPU. This implements model inference on the texture data, and avoids conversion of the texture data into another data type, thereby reducing I/O memory overheads.

Abstract: A vision inspection system includes a sorting platform having an upper surface supporting parts for inspection. An inspection station is positioned adjacent the sorting platform including an imaging device to image the parts in a field of view. A vision inspection controller receives images from the imaging device. The vision inspection controller includes an image histogram tool to pre-process the images to improve contrast of the images by redistributing lightness values of the images based on adaptive histogram equalization processing to generate enhanced images. The vision inspection controller processes the enhanced images based on an image analysis model to determine inspection results for each of the parts. The vision inspection controller has an artificial intelligence learning module operated to customize and configure the image analysis model based on the enhanced images.

Abstract: Training an estimation model using soft labels includes receiving an image. It further includes generating a continuous target map corresponding to the image that includes hard labels and soft labels. A model is trained using the corresponding continuous target map.

Abstract: A computer-implemented unsupervised learning method of training a video feature extractor. The video feature extractor is configured to extract a feature representation from a video sequence. The method uses training data representing multiple training video sequences. From a training video sequence of the multiple training video sequences, a current subsequence; a preceding subsequence preceding the current subsequence; and a succeeding subsequence succeeding the current subsequence are selected. The video feature extractor is applied to the current subsequence to extract a current feature representation of the current subsequence. A training signal is derived from a joint predictability of the preceding and succeeding subsequences given the current feature representation. The parameters of the video feature extractor are updated based on the training signal.

Abstract: A process for reducing time of transmission for single-band, multiple-band or hyperspectral imagery using Machine Learning based compression is disclosed. The process uses Machine Learning to compress single-band, multiple-band and hyperspectral imagery, thereby decreasing the needed bandwidth and storage-capacity requirements for efficient transmission and data storage. The reduced file size for transmission accelerate the communications and reduces the transmission time. This enhances communications systems where there is a greater need for on or near real-time transmission, such as mission critical applications in national security, aerospace and natural resources.

Abstract: Techniques are provided herein for retrieving images that correspond to a target subject matter within a target context. Although useful in a number of applications, the techniques provided herein are particularly useful in contextual product association and visualization. A method is provided to apply product images to a neural network. The neural network is configured to classify the products in the images. The images are associated with a context representing the combination of classified products in the images. These techniques leverage both seller-provided images of products and user-generated content, which potentially includes hundreds or thousands of images of the same or similar products as the seller-provided images. A graphical user interface is configured to permit a user to select the context of interest in which to visualize the products.

Abstract: An on-device training-based user recognition method includes performing on-device training on a feature extractor based on reference data corresponding to generalized users and user data, determining a registration feature vector based on an output from the feature extractor in response to the input of the user data, determining a test feature vector based on an output from the feature extractor in response to an input of test data, and performing user recognition on a test user based on a result of comparing the registration feature vector to the test feature vector.

Abstract: Augmented Denoising Diffusion Implicit Models (“DDIMs”) using a latent trajectory optimization process can be used for image generation and manipulation using text input and one or more source images to create an output image. Noise bias and textual bias inherent in the model representing the image and text input is corrected by correcting trajectories previously determined by the model at each step of a diffusion inversion process by iterating multiple starts the trajectories to find determine augmented trajectories that minimizes loss at each step. The trajectories can be used to determine an augmented noise vector, enabling use of an augmented DDIM and resulting in more accurate, stable, and responsive text-based image manipulation.

Abstract: Some embodiments of the invention provide efficient, expressive machined-trained networks for performing machine learning. The machine-trained (MT) networks of some embodiments use novel processing nodes with novel activation functions that allow the MT network to efficiently define with fewer processing node layers a complex mathematical expression that solves a particular problem (e.g., face recognition, speech recognition, etc.). In some embodiments, the same activation function (e.g., a cup function) is used for numerous processing nodes of the MT network, but through the machine learning, this activation function is configured differently for different processing nodes so that different nodes can emulate or implement two or more different functions (e.g., two or more Boolean logical operators, such as XOR and AND). The activation function in some embodiments is a periodic function that can be configured to implement different functions (e.g., different sinusoidal functions).

Abstract: Customer-agent interactions are often essential to provide services, such as to resolve issues. Methods and systems are provided to enable an artificially intelligent (AI) agent, such as a neural network to annotate a communication, such as a communication comprising a video stream. The AI may determine the subject of an issue and/or an issue to be resolved as a candidate resolution, which may further comprise annotations provided to the video stream. As a benefit the resolution, with annotations may be provided to the agent for subsequent processing and/or the customer.

Abstract: An exemplified methods and systems provides a Volterra filter network architecture that employs a cascaded implementation and a plurality of kernels, a set of which is configured to execute an nth order filter, wherein the plurality of kernels of the nth order filters are repeatedly configured in a plurality of cascading layers of interconnected kernels to form a cascading hierarchical structure that approximates a high-order filter.

Abstract: A system, method and computer program product for accessing content. The method comprises processing at least one image with a classifier, and, in response to the at least one image being processed by the classifier, outputting from the classifier a value indicative of the likelihood that the at least one image belongs to at least one classification. The method also comprises determining whether the at least one image belongs to the at least one classification, based on the value, and accessing predetermined content when it is determined that the at least one image belongs to the at least one classification. Images may be classified by, e.g., genre, musical album, concept, or the like, and, in cases where an image belongs to any such classes, predetermined content (e.g., metadata and/or an audio track) relating thereto is identified and presented to the user.

Abstract: In one embodiment, an apparatus comprises a memory and a processor. The memory is to store visual data associated with a visual representation captured by one or more sensors. The processor is to: obtain the visual data associated with the visual representation captured by the one or more sensors, wherein the visual data comprises uncompressed visual data or compressed visual data; process the visual data using a convolutional neural network (CNN), wherein the CNN comprises a plurality of layers, wherein the plurality of layers comprises a plurality of filters, and wherein the plurality of filters comprises one or more pixel-domain filters to perform processing associated with uncompressed data and one or more compressed-domain filters to perform processing associated with compressed data; and classify the visual data based on an output of the CNN.

Abstract: A manufacturing method of learning data is used for making a neural network perform learning. The manufacturing method of learning data includes a first acquiring step configured to acquire an original image, a second acquiring step configured to acquire a first image as a training image generated by adding blur to the original image, and a third acquiring step configured to acquire a second image as a ground truth image generated by adding blur to the original image. A blur amount added to the second image is smaller than that added to the first image.

Abstract: A system and methods for assessing road surface quality includes a wireless mobile device having a camera, a location receiver, and a road surface classifying computer application and is configured to be mounted on a vehicle. The system has a remote server having a road surface classifying web application, a database, and an interactive map connected to the web application. The mobile device actuates the camera to record videos, extract images from the videos, process the images, classify the images into road conditions, record a location of the images, generate a data packet including an identification of the mobile device and a time stamp of the data packet, and transmit the data packet to the remote server. The remote server stores the data packet in the database. The web application superimposes the time stamp of the data packet, the location, the road conditions, and the images on the interactive map.

Abstract: A method, system and computer program product for segmenting generic foreground objects in images and videos. For segmenting generic foreground objects in videos, an appearance stream of an image in a video frame is processed using a first deep neural network. Furthermore, a motion stream of an optical flow image in the video frame is processed using a second deep neural network. The appearance and motion streams are then joined to combine complementary appearance and motion information to perform segmentation of generic objects in the video frame. Generic foreground objects are segmented in images by training a convolutional deep neural network to estimate a likelihood that a pixel in an image belongs to a foreground object. After receiving the image, the likelihood that the pixel in the image is part of the foreground object as opposed to background is then determined using the trained convolutional deep neural network.

Abstract: The present application provides a method and a system for generating an image sample having a specific feature. The method includes: training a generative adversarial network-based sample generation model, where the generative adversarial network includes a generator and two discriminators: a global discriminator configured to perform global discrimination on an image, and a local discriminator configured to perform local discrimination on a specific feature; and inputting, to a trained generator that serves as a sample generation model, a semantic segmentation image that indicates a location of the specific feature and a corresponding real image not having the specific feature, to obtain a generated image sample having the specific feature.

Abstract: The techniques described herein relate to a systems and methods for a digital asset generation platform. The digital asset generation platform may ingest an ingest input. The digital asset generation platform may utilize a document identification engine corresponding to a first stage of a multi-stage convolutional neural network for identifying document types of documents. The digital asset generation platform may utilize an object detector engine corresponding to a second stage of the multi-stage convolutional neural network for detecting a dynamic mapping in the digital file. The digital asset generation platform may utilize a post-processing engine for classifying the dynamic mapping in the at least one digital file. The digital asset generation platform may dynamically generate a digital asset representative of the document based on the key value data pairs extracted from the dynamic mapping.

Abstract: A method includes accessing a web-based property over a network; storing a plurality of images or videos from the web-based property and associations between the plurality of images or videos and a target audience identifier responsive to the web-based property having a stored association with the target audience identifier; retrieving the plurality of images or videos from the database responsive to each of the plurality of images or videos having stored associations with the target audience identifier; executing a neural network to generate a performance score for each of the plurality of images or videos; calculating a target audience benchmark; executing the neural network to generate a first performance score for a first image or video and a second performance score for a second image or video; comparing the first performance score and the second performance score to the benchmark; and generating a record identifying the first image or video.

Abstract: Helper neural network can play a role in augmenting authentication services that are based on neural network architectures. For example, helper networks are configured to operate as a gateway on identification information used to identify users, enroll users, and/or construct authentication models (e.g., embedding and/or prediction networks). Assuming, that both good and bad identification information samples are taken as part of identification information capture, the helper networks operate to filter out bad identification information prior to training, which prevents, for example, identification information that is valid but poorly captured from impacting identification, training, and/or prediction using various neural networks. Additionally, helper networks can also identify and prevent presentation attacks or submission of spoofed identification information as part of processing and/or validation.

Abstract: A method for ascertaining an explanation map of an image. All those pixels of the image are highlighted which are significant for a classification of the image ascertained with the aid of a deep neural network. The explanation map is being selected in such a way that it selects a smallest possible subset of the pixels of the image as relevant. The explanation map leads to the same classification result as the image when the explanation map is supplied to the deep neural network for classification. The explanation map is selected in such a way that an activation caused by the explanation map does not essentially exceed an activation caused by the image in feature maps of the deep neural network.

Abstract: A neural network in multi-task deep learning paradigm for machine vision includes an encoder that further includes a first, a second, and a third tier. The first tier comprises a first-tier unit having one or more first-unit blocks. The second tier receives a first-tier output from the first tier at one or more second-tier units in the second tier, a second-tier unit comprises one or more second-tier blocks, the third tier receives a second-tier output from the second tier at one or more third-tier units in the third tier, and a third-tier block comprises one or more third-tier blocks. The neural network further comprises a decoder operatively the encoder to receive an encoder output from the encoder as well as one or more loss function layers that are configured to backpropagate one or more losses for training at least the encoder of the neural network in a deep learning paradigm.

Abstract: Methods and systems for automated video segmentation are disclosed. A sequence of video frames having video segments of contextually-related sub-sequences may be received. Each frame may be labeled according to segment and segment class. A video graph may be constructed in which each node corresponds to a different frame, and each edge connects a different pair of nodes, and is associated with a time between video frames and a similarity metric of the connected frames. An artificial neural network (ANN) may be trained to predict both labels for the nodes and clusters of the nodes corresponding to predicted membership among the segments, using the video graph as input to the ANN, and ground-truth clusters of ground-truth labeled nodes. The ANN may be further trained to predict segment classes of the predicted clusters, using the segment classes as ground truths. The trained ANN may be configured for application runtime video sequences.

Abstract: In an approach to augmenting a caption dataset by leveraging a denoising autoencoder to sample and generate additional captions from the ground truth captions, one or more computer processors generate a plurality of new captions utilizing an autoencoder fed with one or more noisy captions, wherein the autoencoder is trained with a dataset comprising a plurality of ground truth captions. The one or more computer processors calculate an importance weight for each new caption in the plurality of generated new captions as compared to a plurality of associated ground truth captions based on a consensus metric. The one or more computer processors train a caption model with the generated plurality of new captions and associated calculated weights.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer readable media for remotely generating modified digital images utilizing an interactive image editing architecture. For example, the disclosed systems receive an image editing request for remotely editing a digital image utilizing an interactive image editing architecture. In some cases, the disclosed systems maintain, via a canvas worker container, a digital stream that reflects versions of the digital image. The disclosed systems determine, from the digital stream utilizing the canvas worker container, an image differential metric indicating a difference between a first version of the digital image and a second version of the digital image associated with the image editing request. Further, the disclosed systems provide the image differential metric to a client device for rendering the second version of the digital image to reflect a modification corresponding to the user interaction.

Abstract: A method in which a convolutional neural network is configured to receive an input data structure including a group of values corresponding to signal samples and to generate a corresponding classification output indicative of a selected one among plural predefined classes. The convolutional neural network includes an ordered sequence of layers, each configured to receive a corresponding layer input data structure including a group of input values, and generate a corresponding layer output data structure including a group of output values by convolving the layer input data structure with at least one corresponding filter including a corresponding group of weights. The layer input data structure of the first layer of the sequence corresponds to the input data structure. The layer input data structure of a generic layer of the sequence different from the first layer corresponds to the layer output data structure generated by a previous layer in the sequence.

Abstract: An electronic device and method of dataset cleaning is provided. The electronic device receives a dataset comprising a plurality of samples, of which a first sample comprises a 2D image of an object of interest and a 3D shape model of the object of interest. The electronic device determines 2D landmarks from the 2D image and extracts 3D landmarks from the 3D shape model. The electronic device computes an error between the determined 2D landmarks and corresponding 2D locations of the extracted 3D landmarks on the 2D image, based on an error metric. Thereafter, the electronic device determines the computed error to be above a threshold. Based on the determination that the computed error is above the threshold, the electronic device updates the dataset by a removal of the first sample from the dataset and trains a neural network on a task of 3D reconstruction, based on the updated dataset.

Abstract: Embodiments of this application disclose a training method using image processing model for processing blurry images. The method includes obtaining a sample pair comprising a clear image and a corresponding blurry image; the sharpness of the clear image being greater than a preset threshold, the sharpness of the blurry image being less than the preset threshold; activating the image processing model to perform sharpness restoration on the blurry image to obtain a restored image; and updating network parameters of a first network and network parameters of a second network in the image processing model according to the restored image and the clear image to obtain a trained image processing model; the network parameters of the first network and the network parameters of the second network meeting a selective sharing condition indicating whether the network parameters between the first network and the second network are shared or independent.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing inputs using an image processing neural network system that includes a spatial transformer module. One of the methods includes receiving an input feature map derived from the one or more input images, and applying a spatial transformation to the input feature map to generate a transformed feature map, comprising: processing the input feature map to generate spatial transformation parameters for the spatial transformation, and sampling from the input feature map in accordance with the spatial transformation parameters to generate the transformed feature map.

Abstract: A system and method for identifying a subject using imaging are provided. In some aspects, the method includes receiving an image depicting a subject to be identified, and applying a trained Disentangled Representation learning-Generative Adversarial Network (DR-GAN) to the image to generate an identity representation of the subject, wherein the DR-GAN comprises a discriminator and a generator having at least one of an encoder and a decoder. The method also includes identifying the subject using the identity representation, and generating a report indicative of the subject identified.

Abstract: A method for an electronic apparatus to perform an operation of an artificial intelligence model includes acquiring resource information for hardware of the electronic apparatus while a plurality of data used for an operation of a neural network model are stored in a memory, the plurality of data respectively having degrees of importance different from each other; obtaining data to be used for the operation of the neural network model among the plurality of data according to the degrees of importance of each of the plurality of data based on the acquired resource information; and performing the operation of the neural network model by using the obtained data.

Abstract: The technology relates to tuning a data translation block (DTB) including a generator model and a discriminator model. One or more processors may be configured to receive training data including an image in a second domain. The image in the second domain may be transformed into a first domain with a generator model. The transformed image may be processed to determine one or more outputs with one or more deep neural networks (DNNs) trained to process data in the first domain. An original objective function for the DTB may be updated based on the one or more outputs. The generator and discriminator models may be trained to satisfy the updated objective function.

Abstract: A computer-implemented method, a computing system, and a computer program product for generating new items compatible with given items may use data associated with a plurality of images and random noise data associated with a random noise image to train an adversarial network including a series of generator networks and a series of discriminator networks corresponding to the series of generator networks by modifying, using a loss function of the adversarial network that depends on a compatibility of the images, one or more parameters of the series of generator networks. The series of generator networks may generate a generated image associated with a generated item different than the given items.

Abstract: A system for testing and/or training the vision of a user is disclosed herein. The system includes at least one camera, a visual display device having an output screen, and a data processing device operatively coupled to the at least one camera and the visual display device. In one embodiment, the data processing device is programmed to determine a head position, head velocity, and/or head speed of a user during a vision test or vision training routine from a plurality of images of the head of the user captured by the at least one camera. In another embodiment, the data processing device is programmed to determine, based upon an input signal received from a user input device, a contrast display setting for a screen background relative to at least one visual target, the contrast display setting enabling the user to gradually adapt to increasing levels of visual stimulation.

Abstract: Provided are an information processing apparatus, an information processing method, and a program capable of accumulating appropriate relearning data. An information processing apparatus includes an input unit that inputs input data to a learned model acquired in advance through machine learning using learning data, an acquisition unit that acquires output data output from the learned model through the input using the input unit, a reception unit that receives correction performed by a user for the output data acquired by the acquisition unit, and a storage controller that performs control for storing, as relearning data of the learned model, the input data and the output data that reflects the correction received by the reception unit in a storage unit in a case where a value indicating a correction amount acquired by performing the correction for the output data is equal to or greater than a threshold value.

Abstract: An electronic apparatus may include a memory that stores first information regarding a plurality of first artificial intelligence models trained to perform image processing differently from each other and second information regarding a second artificial intelligence model trained to identify a type of an image by predicting a processing result of the image by each of the plurality of first artificial intelligence models. The electronic apparatus may further include a processor configured to identify a type of an input image by inputting the input image to the second artificial intelligence model stored in the memory, and process the input image by inputting the input image to one of the plurality of first intelligence models stored in the memory based on the identified type.

Abstract: A method of training a generator G of a Generative Adversarial Network (GAN) includes receiving, by an encoder E, a target data Y; receiving, by the encoder E, an output G(Z) of the generator G, where the generator G generates the output G(Z) in response to receiving a random sample Z and where a discriminator D of the GAN is trained to distinguish which of the G(Z) and the target data Y; training the encoder E to minimize a difference between a first latent space representation E(G(Z)) of the output G(Z) and a second latent space representation E(Y) of the target data Y, where the output G(Z) and the target data Y are input to the encoder E; and using the first latent space representation E(G(Z)) and the second latent space representation E(Y) to constrain the training of the generator G.

Abstract: Mechanisms are provided to provide an improved computer tool for determining and mitigating the presence of adversarial inputs to an image classification computing model. A machine learning computer model processes input data representing a first image to generate a first classification output. A cohort of second image(s), that are visually similar to the first image, is generated based on a comparison of visual characteristics of the first image to visual characteristics of images in an image repository. A cohort-based machine learning computer model processes the cohort of second image(s) to generate a second classification output and the first classification output is compared to the second classification output to determine if the first image is an adversarial image. In response to the first image being determined to be an adversarial image, a mitigation operation by a mitigation system is initiated.

Abstract: A system for recording, storing and processing diagnostic information, including: a computer implementing a computer-readable media including digital data and ground truth; a registry constructed and arranged to store and associate transactions or accesses on the data; and a machine learning system that considers each learning step modification a microtransaction for the data used in that step and which is recorded in the transaction registry. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the methods.

Abstract: According to one embodiment, a training system includes a first generator, a second generator, a third generator, and a trainer. The first generator uses a human body model to generate a first image. The human body model models a human body and is three-dimensional and virtual. The second generator generates a teacher image by annotating body parts of the human body model in the first image. The third generator generates a second image including noise by performing, on the first image, at least one selected from first processing, second processing, third processing, fourth processing, or fifth processing. The trainer uses the second image and the teacher image to train a first model.

Abstract: A computer that identifies a fake image is described. During operation, the computer receives an image. Then, the computer performs analysis on the image to determine a signature that includes multiple features. Based at least in part in the determined signature, the computer classifies the image as having a first signature associated with the fake image or as having a second signature associated with a real image, where the first signature corresponds to a finite resolution of a neural network that generated the fake image, a finite number of parameters in the neural network that generated the fake image, or both. For example, the finite resolution may correspond to floating point operations in the neural network. Moreover, in response to the classification, the computer may perform a remedial action, such as providing a warning or a recommendation, or performing filtering.

Abstract: A recognition apparatus and a training method are provided. The recognition apparatus includes a memory configured to store a neural network including a previous layer of neurons, and a current layer of neurons that are activated based on first synaptic signals and second synaptic signals, the first synaptic signals being input from the previous layer, and the second synaptic signals being input from the current layer. The recognition apparatus further includes a processor configured to generate a recognition result based on the neural network. An activation neuron among the neurons of the current layer generates a first synaptic signal to excite or inhibit neurons of a next layer, and generates a second synaptic signal to inhibit neurons other than the activation neuron in the current layer.

Abstract: In accordance with implementations of the subject matter described herein, a solution for fine-grained image recognition is proposed. This solution includes extracting a global feature of an image using a first sub-network of a first learning network; determining a first attention region of the image based on the global feature using a second sub-network of the first learning network, the first attention region including a discriminative portion of an object in the image; extracting a first local feature of the first attention region using a first sub-network of a second learning network; and determining a category of the object in the image based at least in part on the first local feature. Through this solution, it is possible to localize an image region at a finer scale accurately such that a local feature at a fine scale can be obtained for object recognition.