Abstract: A method of data preparation for artificial intelligence models includes receiving data characterizing a first plurality of images. The method further includes annotating a first subset of images of the first plurality of images based at least in part on a first user input to generate annotated first subset of images. The annotating includes labelling one or more features of the first subset of images. The method also includes generating, by a training code, an annotation code, the training code configured to receive the annotated first subset of images as input and output the annotation code. The training and the annotation code includes computer executable instructions. The method also includes annotating, by the annotation code, a second subset of images of the first plurality of images to generate annotated second subset of images, wherein the annotating includes labelling one or more features of the second subset of images.

Abstract: A target recognition method and device based on a MASK RCNN network model are disclosed. The method comprises: determining a multi-stage network as a basic network; selecting at least one intermediate layer capable of extracting a feature map from the basic network, and inputting respectively a feature map output by the intermediate layer and a feature map output by an end layer of the basic network to corresponding MASK RCNN recognition networks to construct a network model based on the MASK RCNN, wherein the feature map output by the intermediate layer and the feature map output by the end layer have different sizes; training the MASK RCNN recognition networks with a data set and stopping training until a preset training end condition is satisfied; and recognizing the target using the MASK RCNN recognition networks after trained. This solution is very suitable for small target recognition of a flying UAV.

Abstract: Embodiments described herein include, software, firmware, and hardware logic that provides techniques to perform arithmetic on sparse data via a systolic processing unit. Embodiment described herein provided techniques to skip computational operations for zero filled matrices and sub-matrices. Embodiments additionally provide techniques to maintain data compression through to a processing unit. Embodiments additionally provide an architecture for a sparse aware logic unit.

Abstract: Systems and methods for training image processing models for vehicle data collection by image analysis are provided. An example method involves accessing an image of a field of interest in a vehicle captured by a camera in the vehicle, providing a user interface to, display the image, receive input that defines a region of interest in the image that is expected to convey vehicle information, and receive input that assigns a label to the region of interest that associates the region of interest with an image processing model that is to be trained to extract a type of vehicle information from the region of interest, and contributing the image, labelled with the region of interest and the label associating the region of interest to the image processing model, to a training data library to train the image processing model.

Abstract: Systems and methods for training a model are described herein. In one example, a system for training the model includes a processor and a memory in communication with the processor having a training module. The training module has instructions that cause the processor to determine a contrastive loss using a self-supervised contrastive loss function, adjust, based on the contrastive loss, model weights a visual backbone that generated feature maps and/or a textual backbone that generated feature vectors. The training module also has instructions that cause the processor to determine a localized loss using a supervised loss function that compares an image-caption attention map with visual identifiers and adjust, based on the localized loss, the model weights the visual backbone and/or the textual backbone.

Abstract: Provided is an apparatus for classifying an attribute of an image object, including: a first memory configured to store target object images that are indexed; a second memory configured to store target object images that are un-indexed; and an object attribute classification module configured to perform learning on the un-indexed target object images to construct a classifier for classifying a detailed attribute of target object, and finely adjust the classifier on the basis of the indexed target object images.

Abstract: A system and method for displaying and navigating breast tissue is configured for or includes obtaining a plurality of 2D and/or 3D images of a patient's breast; generating a synthesized 2D image of the breast from the obtained images; displaying the synthesized 2D image; receiving a user command, or otherwise detecting through a user interface, a user selection or other indication of an object or region in the synthesized 2D image; and displaying at least a portion of one or more images from the plurality, including a source image and/or most similar representation of the user selected or indicated object or region.

Abstract: Aspects of the present disclosure include methods for identifying one or more components of a sample in a flow stream using a dynamic algorithm (e.g., a machine learning algorithm). Methods according to certain embodiments include detecting light from a sample having particles in a flow stream, generating a data signal of parameters of the particles from the detected light, generating an image based on the data signal, comparing the image with one or more image classification parameters and classifying one or more components of the image using a dynamic algorithm that updates the image classification parameters based on the classified components in the image. Systems and integrated circuit devices programmed for practicing the subject methods, such as on a flow cytometer, are also provided.

Abstract: This disclosure describes an activity recognition system for asymmetric (e.g. left- and right-handed) activities that leverages the symmetry intrinsic to most human and animal bodies. Specifically, described is 1) a human activity recognition system that only recognizes handed activities but is inferenced twice, once with input flipped, to identify both left- and right-handed activities and a training method for learning-based implementations of the aforementioned system that flips all training instances (and associated labels) to appear left-handed and in doing so, balances the training dataset between left- and right-handed activities.

Abstract: Certain aspects of the present disclosure provide techniques for extracting data from a document. An example method generally includes identifying a bounding polygon of the region from an electronic image of the document and extracting data from within the bounding polygon of the region. The method further includes generating revised extracted data based on the extracted data, and combining the revised extracted data with other data extracted from the electronic image of the document to generate input data for a data processing application.

Abstract: A method includes: obtaining a plurality of clinical red blood cell images, dividing red blood cells of different shapes at different positions in each of the red blood cell images into a plurality of submasks, and synthesizing the submasks corresponding to each of the red blood cell images to generate one mask to obtain a plurality of masks corresponding to the red blood cell images; collecting shape data of a plurality of red blood cells from the masks to obtain a training data set, calculating a segmentation boundary of each red blood cell in the training data set, and establishing a red blood cell shape data set based on the segmentation boundary of each red blood cell; collecting distribution data of each red blood cell in the red blood cell shape data set; and synthesizing the red blood cell shape data set into a plurality of red blood cell images.

Abstract: A flexible inspection system includes a robot with a plurality of scanners and a robot controller. The robot controller is configured to receive a vehicle inspection protocol (VIP) for a vehicle being assembled on an assembly line. The VIP includes checkpoints to be scanned on the vehicle and the checkpoints correspond to components installed on the vehicle and connections between components installed on the vehicle. The robot controller commands the robot to move the plurality of scanners per the VIP such that the checkpoints are scanned. A characteristic of each checkpoint is recorded and compared to a reference characteristic such that a pass or no-pass determination of each checkpoint is provided. A vehicle inspection report with the pass/no-pass determinations is provided to an operator such that operator inspections and/or repairs of the checkpoints are made.

Abstract: Methods, apparatus, systems, and articles of manufacture are disclosed to decode receipts based on neural graph architecture. An example apparatus for decoding receipts includes, vertex feature representation circuitry to extract features from optical-character-recognition (OCR) words, polar coordinate circuitry to: calculate polar coordinates of the OCR words based on respective ones of the extracted features, graph neural network circuitry to generate an adjacency matrix based on the extracted features, post-processing circuitry to traverse the adjacency matrix to generate cliques of OCR processed words, and output circuitry to generate lines of text based on the cliques of OCR processed words.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer-readable media for generating a response to a question received from a user during display or playback of a video segment by utilizing a query-response-neural network. The disclosed systems can extract a query vector from a question corresponding to the video segment using the query-response-neural network. The disclosed systems further generate context vectors representing both visual cues and transcript cues corresponding to the video segment using context encoders or other layers from the query-response-neural network. By utilizing additional layers from the query-response-neural network, the disclosed systems generate (i) a query-context vector based on the query vector and the context vectors, and (ii) candidate-response vectors representing candidate responses to the question from a domain-knowledge base or other source.

Abstract: A system for optimizing a machine learning model. The machine learning model generates predictions based on at least one input feature vector, each input feature vector having one or more vector values; and an optimization module with a processor and an associated memory, the optimization module being configured to: create at least one slice of the predictions based on at least one vector value, determine at least one optimization metric of the slice that is based on at least a total number of predictions for the vector value, and optimize the machine learning model based on the optimization metric.

Abstract: A system for monitoring vehicle traffic may include a camera positioned to capture images within a license plate detection zone, the images may represent license plates of vehicles. The system may include an electronic device identification sensor that detects and stores electronic device identifiers of electronic devices located within an electronic device detection zone, and a computing system that detects, using the images, a license plate ID of a vehicle, compare the license plate ID of the vehicle to a database of trusted vehicle license plate IDs, identifies the vehicle as a suspicious vehicle, the identification based at least in part on the comparison of the license plate ID of the vehicle to the database of trusted vehicle license plate IDs, and correlates the license plate ID of the vehicle with at least one of the plurality of stored electronic device identifiers.

Abstract: Data is received that includes a feed of images of a plurality of objects passing in front of an inspection camera module forming part of a quality assurance inspection system. A representation is generated for each image using a first machine learning model. One or more second machine learning models are then used to analyze each image using the corresponding representation. The analyses can be provided to a consuming application or process for quality assurance analysis.

Abstract: 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.

Abstract: Aligning multiple 3D images of an object can be difficult when the representative datasets (images) are large. An exemplary aspect of this technology teaches a technique to subdivide the images and use the alignments between the subdivided images to determine the alignment between the complete datasets.

Abstract: A convolutional neural network includes: convolution layers and a merging layer. At least one convolution layer includes a crossbar circuit having input bars, output bars and weight assignment elements that assign weights to input signals. The crossbar circuit performs a convolution operation in an analog region with respect to input data including the input signal by adding the input signals at each output bar. The input data includes feature maps. The crossbar circuit includes a first crossbar circuit for performing the convolution operation with respect to a part of the feature maps and a second crossbar circuit for performing the convolution operation with respect to another part of feature maps. The merging layer merges convolution operation results of the first and second crossbar circuits.