Abstract: An image manipulation system for generating modified images using a generative adversarial network (GAN) trains GANs using domain changes, aligns input images with generated images, classifies and associates target images based on a symmetry, and uses a modified discriminator structure. A method for domain changes includes generating, using a pre-trained GAN trained on a plurality of first target images, a plurality of images, and determining a feature for each of the plurality of images. The method further includes determining the feature for each of a plurality of second target images and matching, based on the feature, second target images of the plurality of second target images with the plurality of images. The method further includes training a discriminator of the pre-trained GAN with the second target images and the plurality of images.

Abstract: Implementations are directed to training a machine learning model that, once trained, is used in performance of robotic grasping and/or other manipulation task(s) by a robot. The model can be trained using simulated training examples that are based on simulated data that is based on simulated robot(s) attempting simulated manipulations of various simulated objects. Portion(s) of the model can also be trained based on real training examples that are based on data from real-world physical robots attempting manipulations of various objects. The simulated training examples can be utilized to train the model to predict an output that can be utilized in a particular task—and the real training examples used to adapt at least a portion of the model to the real-world domain can be tailored to a distinct task. In some implementations, domain-adversarial similarity losses are determined during training, and utilized to regularize at least portion(s) of the model.

Abstract: Systems for sorting seeds are disclosed, as well as batches of seeds that have been sorted using the systems.

Abstract: A base pathway of a computerized two-pathway video action recognition model is trained using a plurality of labeled video samples. The base pathway is trained using a plurality of unlabeled video samples at a first framerate. An auxiliary pathway of the computerized two-pathway video action recognition model is trained using a plurality of the unlabeled video samples at a second framerate, the second framerate being slower than the first framerate, wherein the training of the base pathway and the training of the auxiliary pathway result in a trained computerized two-pathway video action recognition model. A candidate video is categorized using the trained computerized two-pathway video action recognition model and the categorized candidate video is stored in a computer-accessible video database system for information retrieval.

Abstract: Methods and systems for training a model include combining data from multiple datasets, the datasets having different respective label spaces. Relationships between labels in the different label spaces are identified. A unified neural network model is trained, using the combined data and the identified relationships to generate a unified model, with a class relational binary cross-entropy loss.

Abstract: A computer system trains a federated learning model. A federated learning model is distributed to a plurality of computing nodes, each having a set of local training data comprising labeled data samples. Statistical data is received from each computing node that indicates the node's count of data samples for each label, and is analyzed to identify one or more computing nodes having local training data in which a label category is underrepresented beyond a threshold value with respect to data samples. Additional data samples labeled with the underrepresented labels are provided, and the computing nodes perform training. Results of training are received and are processed to generate a trained global model. Embodiments of the present invention further include a method and program product for training a federated learning model in substantially the same manner described above.

Abstract: A prediction system harvests geo-tagged ground-level images through one or more algorithms. The system receives point of interest data representing structures or events and tags the geo-tagged ground-level images with a probability describing a classification. The system tags point of interest data with a hierarchical genre classification and encodes the tagged geo-tagged ground-level images as vectors to form nodes and edges in a proximity graph. The system encodes tagged points of interest data as similarity vectors to render more nodes and more edges on the proximity graph associated with the tagged geo-tagged ground-level images nodes by calculated semantic distances. The system splits the proximity graph into a training subgraph and a testing subgraph and trains a neural network by aggregating and sampling information from neighboring nodes within the training subgraph graph and validates through the testing subgraph. Training ends when a loss measurement is below a threshold.

Abstract: Techniques are provided for computing systems, methods and computer program products to produce efficient image-to-image translation by adapting unpaired datasets for supervised learning. A first model (a powerful model) may be defined and conditioned using unsupervised learning to produce a synthetic paired dataset from the unpaired dataset, translating images from a first domain to a second domain and images from the second domain to the first domain. The synthetic data generated is useful as ground truths in supervised learning. The first model may be conditioned to overfit the unpaired dataset to enhance the quality of the paired dataset (e.g. the synthetic data generated). A run-time model such as for a target device is trained using the synthetic paired dataset and supervised learning. The run-time model is small and fast to meet the processing resources of the target device (e.g. a personal user device such as a smart phone, tablet, etc.

Abstract: Provided are an image processing apparatus, an image processing method, and a program that can suppress an error in the segmentation of a medical image. An image processing apparatus includes: a segmentation unit (42) that applies deep learning to perform segmentation which classifies a medical image (200) into a specific class on the basis of a local feature of the medical image; and a global feature classification unit (46) that applies deep learning to classify the medical image into a global feature which is an overall feature of the medical image. The segmentation unit shares a weight of a first low-order layer which is a low-order layer with a second low-order layer which is a low-order layer in the global feature classification unit.

Abstract: Methods and systems for generating a simulated image for a specimen are provided. One system includes one or more computer subsystems and one or more components executed by the one or more computer subsystems. The one or more components include a generative adversarial network (GAN), e.g., a conditional GAN (cGAN), trained with a training set that includes portions of design data for one or more specimens designated as training inputs and corresponding images of the one or more specimens designated as training outputs. The one or more computer subsystems are configured for generating a simulated image for a specimen by inputting a portion of design data for the specimen into the GAN.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a classification neural network.

Abstract: The present disclosure relates to systems and methods for image processing. The methods may include obtaining imaging data of a subject, generating a first image based on the imaging data, and generating at least two intermediate images based on the first image. At least one of the at least two intermediate images may be generated based on a machine learning model. And the at least two intermediate images may include a first intermediate image and a second intermediate image. The first intermediate image may include feature information of the first image, and the second intermediate image may have lower noise than the first image. The methods may further include generating, based on the first intermediate image and at least one of the first image or the second intermediate image, a target image of the subject.

Abstract: Apparatus, methods, and systems that operate to provide interactive streaming content identification and processing are disclosed. An example apparatus includes a classifier to determine an audio characteristic value representative of an audio characteristic in audio; a transition detector to detect a transition between a first category and a second category by comparing the audio characteristic value to a threshold value among a set of threshold values, the set of threshold values corresponding to the first category and the second category; and a context manager to control a device to switch from a first fingerprinting algorithm to a second fingerprinting algorithm different than the first fingerprinting algorithm, responsive to the detected transition between the first category and the second category.

Abstract: Disclosed herein is an image preprocessing/analysis apparatus using machine learning-based artificial intelligence. The image preprocessing apparatus includes a computing system, and the computing system includes: a processor; a communication interface configured to receive an input image; and an artificial neural network configured to generate first and second preprocessing conditions through inference on the input image. The processor includes a first preprocessing module configured to generate a first preprocessed image and a second preprocessing module configured to generate a second preprocessed image.

Abstract: A method for converting a lineless table into a lined table includes associating a first set of tables with a second set of tables to form a set of multiple table pairs that includes tables with lines and tables without lines. A conditional generative adversarial network (cGAN) is trained, using the table pairs, to produce a trained cGAN. Using the trained cGAN, lines are identified for overlaying onto a lineless table. The lines are overlaid onto the lineless table to produce a lined table.

Abstract: The present disclosure provides a method for selecting an annotated sample. The method includes: determining a first attribute and a second attribute of a sample characteristic; in which the first attribute is a characteristic attribute of the sample characteristic in a source field sample set, and the second attribute is a characteristic attribute of the sample characteristic in a target field sample set; and determining a target annotated sample from a plurality of candidate annotated samples of the source field sample set according to the first attribute and the second attribute; in which the target annotated sample is configured to train a classification model, the classification model includes a model for determining an emotion polarity by analyzing an input sample to be classified.

Abstract: In one embodiment, a method includes by a computing device, detecting a sensory input, identifying, using a machine-learning model, one or more attributes associated with the machine-learning model, wherein the attributes are identified based on the sensory input in accordance with the model's training, and presenting the attributes as output. The identifying may be performed at least in part by an inference engine that interacts with the model. The sensory input may include an input image received from a camera, and the model may identify the attributes based on an input object in the input image in accordance with the model's training. The model may include a convolutional neural network trained using training data that associates training sensory input with the attributes. The training sensory input may include a training image of a training object, and the input object may be classified in the same class as the training object.

Abstract: The technology employs a patch-based multi-scale Transformer (300) that is usable with various imaging applications. This avoids constraints on image fixed input size and predicts the quality effectively on a native resolution image. A native resolution image (304) is transformed into a multi-scale representation (302), enabling the Transformer's self-attention mechanism to capture information on both fine-grained detailed patches and coarse-grained global patches. Spatial embedding (316) is employed to map patch positions to a fixed grid, in which patch locations at each scale are hashed to the same grid. A separate scale embedding (318) is employed to distinguish patches coming from different scales in the multiscale representation. Self-attention (508) is performed to create a final image representation. In some instances, prior to performing self-attention, the system may prepend a learnable classification token (322) to the set of input tokens.

Abstract: A method for detecting a grasping position of a robot in grasping a target object includes: collecting a target RGB image and a target Depth image of the target object at different view angles; inputting each of the target RGB image to a target object segmentation network for calculation to obtain an RGB pixel region of the target object in the target RGB image and a Depth pixel region of the target object; inputting the RGB pixel region to an optimal grasping position generation network to obtain an optimal grasping position for grasping the target object; inputting the Depth pixel region of the target object and the optimal grasping position to a grasping position quality evaluation network to calculate a score of the optimal grasping position; and selecting an optimal grasping position corresponding to a highest score as a global optimal grasping position of the robot.

Abstract: An image recognition method, a training system for an object recognition model and a training method for an object recognition model are provided. The image recognition method includes the following steps. At least one original sample image of an object in a field and an object range information and an object type information in the original sample image are obtained. At least one physical parameter is adjusted to generate plural simulated sample images of the object. The object range information and the object type information of the object in each of the simulated sample images are automatically marked. A machine learning procedure is performed to train an object recognition model. An image recognition procedure is performed on an input image.

Abstract: A method for determining a localization pose of an at least partially automated mobile platform, the mobile platform being equipped to generate ground images of an area surrounding the mobile platform, and being equipped to receive aerial images of the area surrounding the mobile platform from an aerial-image system. The method includes: providing a digital ground image of the area surrounding the mobile platform; receiving an aerial image of the area surrounding the mobile platform; generating the localization pose of the mobile platform with the aid of a trained convolutional neural network, which has a first trained encoder convolutional-neural-network part and a second trained encoder convolutional-neural-network part.

Abstract: An example system includes a processor to receive an image containing an object to be detected. The processor is to detect the object in the image via a binary object detector trained via a self-supervised training on raw and unlabeled videos.

Abstract: Disclosed are techniques for augmenting video datasets for training machine learning algorithms with additional video datasets that are cropped copies of the video datasets. Frames of a received video dataset are divided into a plurality of subframes. For each subframe, a count is tallied corresponding to the cumulative number of pixels changed across the frames of the received video. Counts are compared to determine which subframe includes the most changed pixels across the frames of the video dataset, which is selected as a cropping candidate. The cropping candidate is used to generate copies of the video dataset that are cropped to include at least the cropping candidate and exclude at least some of the remaining portions of each frame of the video dataset that are outside of the cropping candidate. In some embodiments, boundaries of cropping candidates are transformed to generate a plurality of cropped variations of the video dataset.

Abstract: Techniques are disclosed for using and training a descriptor network. An image may be received and provided to the descriptor network. The descriptor network may generate an image descriptor based on the image. The image descriptor may include a set of elements distributed between a major vector comprising a first subset of the set of elements and a minor vector comprising a second subset of the set of elements. The second subset of the set of elements may include more elements than the first subset of the set of elements. A hierarchical normalization may be imposed onto the image descriptor by normalizing the major vector to a major normalization amount and normalizing the minor vector to a minor normalization amount. The minor normalization amount may be less than the major normalization amount.

Abstract: A method for training a generator. The generator is supplied with at least one actual signal that includes real or simulated physical measured data from at least one observation of the first area. The actual signal is translated by the generator into a transformed signal that represents the associated synthetic measured data in a second area. Using a cost function, an assessment is made concerning to what extent the transformed signal is consistent with one or multiple setpoint signals, at least one setpoint signal being formed from real or simulated measured data of the second physical observation modality for the situation represented by the actual signal. Trainable parameters that characterize the behavior of the generator are optimized with the objective of obtaining transformed signals that are better assessed by the cost function. A method for operating the generator, and that encompasses the complete process chain are also provided.

Abstract: Systems and methods for training a conditional generator model are described. Methods receive a sample, and determine a discriminator loss for the received sample. The discriminator loss is based on an ability to determine whether the sample is generated by the conditional generator model or is a ground truth sample. The method determines a secondary loss for the generated sample and updates the conditional generator model based on an aggregate of the discriminator loss and the secondary loss.

Abstract: Systems described herein may use machine classifiers to perform a variety of natural language understanding tasks including, but not limited to multi-turn dialogue generation. Machine classifiers in accordance with aspects of the disclosure may model multi-turn dialogue as a one-to-many prediction task. The machine classifier may be trained using adversarial bootstrapping between a generator and a discriminator with multi-turn capabilities. The machine classifiers may be trained in both auto-regressive and traditional teacher-forcing modes, with the maximum likelihood loss of the auto-regressive outputs being weighted by the score from a metric-based discriminator model. The discriminators input may include a mixture of ground truth labels, the teacher-forcing outputs of the generator, and/or negative examples from the dataset. This mixture of input may allow for richer feedback on the autoregressive outputs of the generator.

Abstract: A training system for training a generator neural network arranged to transform measured sensor data into generated sensor data. The generator network is arranged to receive as input sensor data and a transformation goal selected from a plurality of transformation goals and is arranged to transform the sensor data according to the transformation goal.

Abstract: This document describes change detection criteria for updating sensor-based maps. Based on an indication that a registered object is detected near a vehicle, a processor determines differences between features of the registered object and features of a sensor-based reference map. A machine-learned model is trained using self-supervised learning to identify change detections from inputs. This model is executed to determine whether the differences satisfy change detection criteria for updating the sensor-based reference map. If the change detection criteria is satisfied, the processor causes the sensor-based reference map to be updated to reduce the differences, which enables the vehicle to safely operate in an autonomous mode using the updated reference map for navigating the vehicle in proximity to the coordinate location of the registered object.

Abstract: The present disclosure discloses a method and device for optimizing an object-class model based on a neural network. The method includes: establishing the object-class model based on the neural network, training the object-class model, and realizing classification of target images by using the object-class model that has been trained; and when a new target image is generated, and the new target image is an image corresponding to a new condition of a target and is capable of still being classified into an original classification system, judging a result of identification of the object-class model to the new target image, and if the object-class model is not capable of correctly classifying the new target image, according to the new target image, selecting some of parameters, adjusting the some of parameters, and training to obtain an object-class model that is capable of correctly classifying the new target image.

Abstract: A method for controlling a robotic device is presented. The method includes capturing an image corresponding to a current view of the robotic device. The method also includes identifying a keyframe image comprising a first set of pixels matching a second set of pixels of the image. The method further includes performing, by the robotic device, a task corresponding to the keyframe image.

Abstract: A laser system for amplifying laser light generated from a laser light source and emitting the laser light includes an optical element in an optical path of the laser light and transmits the laser light, a control device to control power to be supplied to the laser system, an imager to capture an image of the optical element, and an image processing circuitry to process the image of the optical element captured by the imager. The image processing circuitry in which reference images of the optical element corresponding to power information relating to the power are prepared in advance includes a comparison unit to compare a captured image of the optical element captured by the imager with a reference image selected by a reference image selection unit, the reference image corresponding to the power information at a time of image capturing by the imager.

Abstract: The image processing apparatus for performing display restriction processing on a captured image captured by a moving robot includes: a task acquisition unit configured to acquire information that corresponds to a property of a task to be executed via the remote operation performed on the moving robot; a target object identification unit configured to identify target objects in the captured image; a restricted target object specification unit configured to specify a target object for which a display restriction is required among the target objects identified in the target object identification unit in accordance with the property of the task to be executed by the moving robot based on the above information; and a display restriction processing unit configured to perform the display restriction processing on a restricted area in the captured image that corresponds to the target object for which display restriction is required.

Abstract: One embodiment of the present invention sets forth a technique for generating simulated training data for a physical process. The technique includes receiving, as input to at least one machine learning model, a first simulated image of a first object, wherein the at least one machine learning model includes mappings between simulated images generated from models of physical objects and real-world images of the physical objects. The technique also includes performing, by the at least one machine learning model, one or more operations on the first simulated image to generate a first augmented image of the first object. The technique further includes transmitting the first augmented image to a training pipeline for an additional machine learning model that controls a behavior of the physical process.

Abstract: A system and method for providing unsupervised domain adaption for spatio-temporal action localization that includes receiving video data associated with a source domain and a target domain that are associated with a surrounding environment of a vehicle. The system and method also include analyzing the video data associated with the source domain and the target domain and determining a key frame of the source domain and a key frame of the target domain. The system and method additionally include completing an action localization model to model a temporal context of actions occurring within the key frame of the source domain and the key frame of the target domain and completing an action adaption model to localize individuals and their actions and to classify the actions based on the video data. The system and method further include combining losses to complete spatio-temporal action localization of individuals and actions.

Abstract: The subject matter described herein includes methods, systems, and computer readable media for mask embedding for realistic high-resolution image synthesis. According to one method for mask embedding for realistic high-resolution image synthesis includes receiving, as input, a mask embedding vector and a latent features vector, wherein the mask embedding vector acts as a semantic constraint; generating, using a trained image synthesis algorithm and the input, a realistic image, wherein the realistic image is constrained by the mask embedding vector; and outputting, by the trained image synthesis algorithm, the realistic image to a display or a storage device.

Abstract: A method of image processing based on a plurality of frames of images, an electronic device, and a storage medium are provided. The method includes: capturing a plurality of frames of original images; obtaining a high dynamic range (HDR) image by performing image synthesis on the plurality of frames of original images; performing artificial intelligent-based denoising on the HDR image to obtain a target denoised image.

Abstract: A system for categorizing seeds of plants into hybrid and non-hybrid categories. Seeds sorted according to the disclosed system are also disclosed.

Abstract: An information processing method for embedding watermark bits into weights of a first neural network includes: obtaining an output of a second neural network by inputting a plurality of input values obtained from a plurality of weights of the first neural network to the second neural network; obtaining second gradients of the respective plurality of input values based on an error between the output of the second neural network and the watermark bits; and updating the weights based on values obtained by adding first gradients of the weights of the first neural network that have been obtained based on backpropagation and the respective second gradients.

Abstract: A computer system uses Learning from Demonstration (LfD) techniques in which a multitude of tasks are demonstrated without requiring careful task set up, labeling, and engineering, and learns multiple modes of behavior from visual data, rather than averaging the multiple modes. As a result, the computer system may be used to control a robot or other system to exhibit the multiple modes of behavior in appropriate circumstances.

Abstract: Methods and apparatus are disclosed for implementing machine learning data augmentation within the die of a non-volatile memory (NVM) apparatus using on-chip circuit components formed on or within the die. Some particular aspects relate to configuring under-the-array or next-to-the-array components of the die to generate augmented versions of images for use in training a Deep Learning Accelerator of an image recognition system by rotating, translating, skewing, cropping, etc., a set of initial training images obtained from a host device. Other aspects relate to configuring under-the-array or next-to-the-array components of the die to generate noise-augmented images by, for example, storing and then reading training images from worn regions of a NAND array to inject noise into the images.

Abstract: A system for training a classification model to be robust against perturbations of multiple perturbation types. A perturbation type defines a set of allowed perturbations. The classification model is trained by, in an outer iteration, selecting a set of training instances of a training dataset; selecting, among perturbations allowed by the multiple perturbation types, one or more perturbations for perturbing the selected training instances to maximize a loss function; and updating the set of parameters of the classification model to decrease the loss for the perturbed instances. A perturbation is determined by, in an inner iteration, determining updated perturbations allowed by respective perturbation types of the multiple perturbation types and selecting an updated perturbation that most increases the loss of the classification model.

Abstract: Action recognition methods are disclosed.

Abstract: A model testing system administers tests to machine learning (ML) models to test the accuracy and the robustness of the ML models. A user interface (UI) associated with the model testing system receives selections of one or more of a plurality of tests to be administered to a ML model under test. Test data produced by one or more of a plurality of testing ML models that correspond to the plurality of tests is provided to the ML model under test based on the selected tests. One or more of a generative patches test, a generative perturbations test and a counterfeit data test can be administered to the ML model under test based on the selections.

Abstract: A system, a method, and a computer program product for determining a sign type of a road sign are disclosed herein. The system comprises a memory configured to store computer-executable instructions and one or more processors configured to execute the instructions to obtain sensor data associated with the road sign, wherein the sensor data comprises data associated with counts of road sign observations, determine one or more features associated with the road sign, based on the obtained sensor data, and determine the sign type of the road sign, based on the one or more features.

Abstract: A method for generating a synthetic image with predefined properties. The method includes the steps of providing first values which characterize the predefined properties of the image that is to be generated and attention weights which characterize a weighting of one of the first values and feeding sequentially the first values and assigned attention weights as input value pairs into an generative automated learning system that includes at least a recurrent connection. An image generation system and a computer program that are configured to carry out the method are also described.

Abstract: A control method includes: deriving an approach location at which the end effector grips an operation object; deriving a scan location for scanning an identifier of the operation object; and based on the approach location and the scan location, creating or deriving a control sequence to instruct the robot to execute the control sequence. The control sequence includes (1) gripping the operation object from a start location; (2) scanning an identifier of the operation object with a scanner located between the start location and a task location; (3) temporarily releasing the operation object from the end effector and regripping the operation object by the end effector to be shifted, at a shift location, when a predetermined condition is satisfied; and (4) moving the operation object to the task location.

Abstract: An information processing method is executed by a computer, and includes: obtaining a first image generated by a multi-pinhole camera; extracting at least one point spread function (PSF) in each of a plurality of regions in the first image; obtaining a second image different from the first image, and reference data used in machine learning for the second image; generating a third image, by convolving each of a plurality of regions in the second image with at least one PSF extracted in a corresponding region of the plurality of regions in the first image; and outputting a pair of the reference data and the third image.

Abstract: The technology relates to assisting large self-driving vehicles, such as cargo vehicles, as they maneuver towards and/or park at a destination facility. This may include a given vehicle transitioning between different autonomous driving modes. Such a vehicles may be permitted to drive in a fully autonomous mode on certain roadways for the majority of a trip, but may need to change to a partially autonomous mode on other roadways or when entering or leaving a destination facility such as a warehouse, depot or service center. Large vehicles such as cargo truck may have limited room to maneuver in and park at the destination, which may also prevent operation in a fully autonomous mode. Here, information from the destination facility and/or a remote assistance service can be employed to aid in real-time semi-autonomous maneuvering.

Abstract: A system including: at least one processor; and at least one memory having stored thereon computer program code that, when executed by the at least one processor, controls the system to: receive a data model identification and a dataset; in response to determining that the data model does not contain a hierarchical structure, perform expectation propagation on the dataset to approximate the data model with a hierarchical structure; divide the dataset into a plurality of channels; for each of the plurality of channels: divide the data into a plurality of microbatches; process each microbatch of the plurality of microbatches through parallel iterators; and process the output of the parallel iterators through single-instruction multiple-data (SIMD) layers; and asynchronously merge results of the SIMD layers.