Abstract: Method of training an object detector for predicting centers of mass of objects projected onto a ground is provided. The method includes steps of: acquiring training images from training data set; inputting each of training images into the object detector to thereby instruct the object detector to perform object detection for the training images and thus generate object detection results including (i) information on predicted bounding boxes, corresponding to one or more ROIs, acquired by predicting each of locations of the objects in the training images and (ii) information on predicted projection points acquired by projecting the centers of mass of the objects onto the ground; and training the object detector by using object detection losses generated by referring to the object detection results and information on ground truths corresponding to the training images.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for reliably performing data compression and data decompression across a wide variety of hardware and software platforms by using integer neural networks. In one aspect, there is provided a method for entropy encoding data which defines a sequence comprising a plurality of components, the method comprising: for each component of the plurality of components: processing an input comprising: (i) a respective integer representation of each of one or more components of the data which precede the component in the sequence, (ii) an integer representation of one or more respective latent variables characterizing the data, or (iii) both, using an integer neural network to generate data defining a probability distribution over the predetermined set of possible code symbols for the component of the data.

Abstract: A system and method for detecting the origin of wooden planks in a sawmill is provided. The method scans surfaces of processed planks and, with the help of an AI algorithm comprising a deep-learning algorithm, determines the origin of said planks based on analysed parameters of the planks. The parameters used in the analysis are mainly properties of tool marks and the resulting analysis provides tools and equipment used. The deep learning algorithm may be in a self-learning mode or in a supervised training mode.

Abstract: Actors may be detected and tracked within a scene using multiple imaging devices provided in a network that are aligned with fields of view that overlap at least in part. Processors operating on the imaging devices may evaluate the images using one or more classifiers to recognize body parts within the images, and to associate the body parts with a common actor within the scene. Each of the imaging devices may generate records of the positions of the body parts and provide such records to a central server, that may correlate body parts appearing within images captured by two or more of the imaging devices and generate a three-dimensional model of an actor based on positions of the body parts. Motion of the body parts may be tracked in subsequent images, and the model of the actor may be updated based on the motion.

Abstract: Example device meters disclosed herein include a signature reporter to report, to a data processor, media signatures of a first type to monitor media presented by a media device Disclosed example device meters also include a signature generator to (i) generate the media signatures of the first type, (ii) generate media signatures of a second type, different from the first type, to continue monitoring the media presented by the media device after receipt, from the data processor, of an indication that a first media signature of the first type is associated with first reference media, and (iii) in response to detection at the device meter of a change in a source of the media presented by the media device, revert to generation of the media signatures of the first type to monitor the media presented by the media device.

Abstract: Example local devices disclosed herein include memory including a set of reference fingerprints corresponding to media, the set of reference fingerprints from a remote device different from the local device and one or more processor circuits to execute machine readable instructions to generate a monitored fingerprint of the media presented at a location and compare the monitored fingerprint to at least some of the set of reference fingerprints from the remote device. Additionally, the one or more processor circuits are to determine an amount of time that has passed since the media started and after a match between the monitored fingerprint and one or more reference fingerprints of the set of reference fingerprints, cause transmission of audience measurement information to identify the media, the audience measurement information including data indicative of the amount of time that has passed since the media started.

Abstract: Methods, systems, and apparatus for training a machine-learned model using satellite imagery and physical river gauge data as ground-truth information. Methods include receiving, from a user in a graphical user interface presented on a user device, a depth request for depth information at a geolocation. At least two satellite images are received, including the geolocation where a difference in respective capture times of each of the satellite images is within a threshold. The satellite images for the geolocation are provided to a machine-learned river gauge model. The machine-learned river gauge model determines depth information for the geolocation utilizing the satellite images, and provides, to the user in the graphical user interface, the depth information at the geolocation.

Abstract: Methods and systems for meta-learning are described for automating learning of child tasks with a single neural network. The order in which tasks are learned by the neural network can affect performance of the network, and the meta-learning approach can use a task-level curriculum for multi-task training. The task-level curriculum can be learned by monitoring a trajectory of loss functions during training. The meta-learning approach can learn to adapt task loss balancing weights in the course of training to get improved performance on multiple tasks on real world datasets. Advantageously, learning to dynamically balance weights among different task losses can lead to superior performance over the use of static weights determined by expensive random searches or heuristics. Embodiments of the meta-learning approach can be used for computer vision tasks or natural language processing tasks, and the trained neural networks can be used by augmented or virtual reality devices.

Abstract: A method includes the following steps: pre-processing chest X-ray films to obtain initial X-ray film images that meets format requirements; screening the initial X-ray film images to detect whether they are posteroanterior chest images; inputting the posteroanterior chest images into a binary classification model of the deep convolutional neural network for negative and positive classification; inputting the images presenting positive results into a detection model of the deep convolutional neural network to detect a disease type and label an outline of a lesion area in each image; and displaying the disease type and lesion area corresponding to the image.

Abstract: Provided are various mechanisms and processes for automatic computer vision-based defect detection using a neural network. A system is configured for receiving historical datasets that include training images corresponding to one or more known defects. Each training image is converted into a corresponding matrix representation for training the neural network to adjust weighted parameters based on the known defects. Once sufficiently trained, a test image of an object that is not part of the historical dataset is obtained. Portions of the test image are extracted as input patches for input into the neural network as respective matrix representations. A probability score indicating the likelihood that the input patch includes a defect is automatically generated for each input patch using the weighted parameters. An overall defect score for the test image is then generated based on the probability scores to indicate the condition of the object.

Abstract: A neural network is trained for use in a substrate thickness measurement system by obtaining ground truth thickness measurements of a top layer of a calibration substrate at a plurality of locations, each location at a defined position for a die being fabricated on the substrate. A plurality of color images of the calibration substrate are obtained, each color image corresponding to a region for a die being fabricated on the substrate. A neural network is trained to convert color images of die regions from an in-line substrate imager to thickness measurements for the top layer in the die region. The training is performed using training data that includes the plurality of color images and ground truth thickness measurements with each respective color image paired with a ground truth thickness measurement for the die region associated with the respective color image.

Abstract: A computer-implemented method for training a software infrastructure based on machine-learning techniques to analyse data obtained from a instrumental examination of objects of a predetermined type, where each of the objects has been obtained by splitting a product into smaller pieces, wherein the software infrastructure receives, for each object in a training set, training input data comprising the data obtained from the instrumental examination and training output data comprising information on the characteristics of interest of the training object, wherein the information on the characteristics of interest is, at least in part, information that has been obtained from the results of a tomographic examination of the product from which the training object was obtained, and wherein the software infrastructure processes, through its own training unit, the training input data and the training output data for each training object in order to set internal processing parameters for the software infrastructure which

Abstract: Knowledge distillation method for fracture detection includes obtaining medical images including region-level labeled images, image-level diagnostic positive images, and image-level diagnostic negative images, in chest X-rays; performing a supervised pre-training process on the region-level labeled images and the image-level diagnostic negative images to train a neural network to generate pre-trained weights; and performing a semi-supervised training process on the image-level diagnostic positive images using the pre-trained weights. A teacher model is employed to produce pseudo ground-truths (GTs) on the image-level diagnostic positive images for supervising training of a student model, and the pseudo GTs are processed by an adaptive asymmetric label sharpening (AALS) operator to produce sharpened pseudo GTs to provide positive detection responses on the image-level diagnostic positive images.

Abstract: A method for training an image classification model according to an embodiment includes training a feature extractor and a rotation angle classifier to predict a rotation angle of each of unlabeled first training images, training the image classification model to predict a label and rotation angle of each of labeled second training images, but predict a uniform label even though an actual rotation angle of each of the second training images is changed, generating a pseudo label based on a training image that satisfy a preset condition among unlabeled candidate images, and training the image classification model to predict a rotation angle of each of the third training images, and predict a label of each of the third training images based on the pseudo label, but predict a uniform label even though an actual rotation angle of each of the third training images is changed.

Abstract: The present application discloses a method for inspecting a battery tab, the method including: obtaining a sectional view of a plurality of layers of tabs of a battery; identifying and analyzing the sectional view to obtain a plurality of connected domains, where each connected domain includes one tab or a plurality of tabs that are bonded with each other; determining, based on positions and a number of intersection points of tab bonding in each connected domain, a number of layers of tabs corresponding to the connected domain; calculating a total number of layers of the plurality of layers of tabs in the sectional view based on the number of layers of tabs corresponding to the connected domain; and determining, based on the total number of layers of tabs and a preset real number of layers of tabs, whether the plurality of layers of tabs are folded.

Abstract: Disclosed are an image processing method, an image processing device, a neutral network and a training method thereof, and a storage medium. The image processing method includes: obtaining an input image; performing a segmentation process on the input image via a first encoding-decoding network, to obtain a first output feature map and the first segmented image; concatenating the first output feature map with at least one selected from the group consisting of the input image and the first segmented image, to obtain an input of the second encoding-decoding network; and performing a segmentation process on the input of the second encoding-decoding network via a second encoding-decoding network, to obtain the second segmented image. And the first encoding-decoding network and the second encoding-decoding network forms a neural network.

Abstract: The present disclosure is directed to systems and methods for classifying biomedical images. A feature classifier may generate a plurality of tiles from a biomedical image. Each tile may correspond to a portion of the biomedical image. The feature classifier may select a subset of tiles from the plurality of tiles by applying an inference model. The subset of tiles may have highest scores. Each score may indicate a likelihood that the corresponding tile includes a feature indicative of the presence of the condition. The feature classifier may determine a classification result for the biomedical image by applying an aggregation model. The classification result may indicate whether the biomedical includes the presence or lack of the condition.

Abstract: An information processing apparatus comprises a control unit configured to set a shift amount based on a bit width of data, for each layer of a network including a plurality of layers, a plurality of MAC (multiply-accumulate) units configured to execute MAC operations on a plurality of data and a plurality of filter coefficients of the layer, a plurality of shift operation units configured to shift a plurality of MAC operation results obtained by the plurality of MAC units based on the shift amount, and an adding unit configured to calculate a total sum of the plurality of MAC operation results shifted by the plurality of shift operation units.

Abstract: Object recognition scanning systems and methods for implementing artificial intelligence based item determination are disclosed herein. Example object recognition scanning systems and methods include imaging, by an imager having a field of view (FOV) extending over a scanning area, one or more items within the FOV, and receiving image data of an item imaged by the imager during a scanning session. A trained object recognition model, taking the image data as input, determines a product identification probability for the item. The object recognition scanning systems and methods include executing one of: (a) a first decoding sequence including, (b) a second decoding sequence, or (c) a mismatch detection sequence.

Abstract: A method for an image processing circuit includes steps of: receiving a fingerprint image; performing a low-pass filtering on the fingerprint image to remove a moiré signal on the fingerprint image, to generate a filtered image; and performing a data binning on the filtered image to generate an output image.