196 results about "Ground truth" patented technology

This invention discloses an intelligent identification and positioning method for ocean internal waves based on multi-source remote sensing imagery, belonging to the field of marine remote sensing technology. The key technical solutions include the following steps: determining the geographical location and time range of the target sea area; collecting multi-temporal high-resolution remote sensing image data from synthetic aperture radar (SAR) and optical satellite imagery; labeling the internal wave features of the preprocessed multi-temporal high-resolution remote sensing image data to generate ground truth images; using a trained deep convolutional neural network model to perform pixel-level analysis on the input remote sensing images to extract ocean internal wave features and generate binarized images of the internal wave features; and converting the extracted binarized images of the internal wave features into corresponding latitude and longitude locations based on the geographical location information of the original remote sensing images. This achieves high-precision geographic positioning of internal waves, is suitable for marine research and high-precision environmental monitoring, and has broad application prospects.

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

A method for generating, by an encoder-based model, a three-dimensional (3D) representation of a two-dimensional (2D) image is provided. The encoder-based model is trained to infer the 3D representation using a synthetic training data set generated by a pre-trained model. The pre-trained model is a 3D generative model that produces a 3D representation and a corresponding 2D rendering, which can be used to train a separate encoder-based model for downstream tasks like estimating a triplane representation, neural radiance field, mesh, depth map, 3D key points, or the like, given a single input image, using the pseudo ground truth 3D synthetic training data set. In a particular embodiment, the encoder-based model is trained to predict a triplane representation of the input image, which can then be rendered by a volume renderer according to pose information to generate an output image of the 3D scene from the corresponding viewpoint.

In various examples, systems and method are provided for generation of ground truth gaze data for training in-cabin monitoring systems and applications. A gaze target projector mounted to a known position inside a cabin may be used to project a gaze target onto an interior surface of the cabin. Because a beam of light may be used to produce the projected gaze target, the projected gaze target may be displayed at a projection point on the surface of the cabin interior, even if the surface at the projection point is curved, small, or an irregular shape. Three-dimensional coordinates of a projected gaze target in the cabin coordinate system may be determined and used to label image data that is captured as a projected gaze target is selectively projected onto an interior surface of the cabin and a test occupant's gaze is directed at the projected gaze target.

The objective is to provide image diagnostic technology that utilizes machine learning models. [Solution] One aspect of the present disclosure relates to a machine learning model trained using training data that includes first ultrasonic image data based on a received signal received by an ultrasonic transducer, first ground truth data which is first region information associated with a detection target of the first ultrasonic image data, and second ground truth data which is either first location information associated with a detection target of the first ultrasonic image data, or second region information based on the first location information.

Aspects of the subject disclosure may include, for example, a method performed by a processing system; the method includes receiving a plurality of values of key performance indicators (KPIs) relating to performance of a cell on a communication network. The plurality of values of the KPIs includes labeled training data for training a machine learning (ML) model for the performance of the cell. The method further includes iteratively executing, using the labeled training data, a training procedure for the ML model; and testing the trained ML model. The labeled training data corresponds to ground truth data that may include a training data set, a validation data set and a test data set. The trained ML model, when deployed on a communication network, receives as input near-real time data regarding the performance of the cell and provides as output predictions of the performance of the cell. Other embodiments are disclosed.

This disclosure provides methods and apparatus for generating training data, training machine learning models, and analyzing biological samples. Generating training data includes: acquiring a first image of a biological sample; generating a first set of annotations for the first image based on a second image of the biological sample stained by a quantitative method that converts antibody / antigen complexes into dots; and outputting the first image and a benchmark truth including the first set of annotations as data for training a first machine learning model (“ML”) to analyze the biological sample. Furthermore, the generated (output) training data can then be used to train the ML model, and the trained ML model can be used to analyze the biological sample.

This invention relates to a method for continuous motion prediction in patients with cerebral palsy. The method includes: acquiring sEMG data, IMU data, and ground truth joint angles of the subjects and constructing a dataset; dividing the dataset into training, validation, and test sets for each subject according to time sequence; standardizing the collected EMG and IMU data for each subject to obtain standardized sequence data; inputting the standardized sequence data of each subject into a trained prediction network, which includes parallel TCN and GAT modules. The TCN module extracts IMU temporal features, and the GAT module extracts sEMG spatial features. The two types of features are fused through a multimodal fusion module and then input into a BiLSTM network for bidirectional temporal modeling to capture the bidirectional temporal dependence of movements. Finally, a regression output module outputs the standardized predicted angles. Compared with existing technologies, this invention has advantages such as achieving continuous and accurate motion prediction for patients with cerebral palsy and enhancing cross-subject generalization ability.

This application discloses a model training method, an image processing method, and an electronic device. It belongs to the field of image processing technology. An embodiment of the method includes: acquiring a training dataset; wherein the training dataset includes multiple triples, each triple including a source image, a style reference image, and a ground truth image, the style reference image and the ground truth image having the same color tone style; training an initial model based on the training dataset to obtain a style transfer model; wherein, during the training process, a first loss function, a second loss function, and a third loss function are used for supervision, the first loss function indicating pixel-level reconstruction loss, the second loss function indicating style-aware loss, and the third loss function indicating key region preservation loss.

This invention discloses a short-term precipitation prediction method based on big data. The method includes radar data acquisition, radar echo image interference suppression, spatiotemporal feature encoding construction, differentiated adversarial loss design, short-term precipitation prediction model construction, and short-term precipitation prediction. This invention belongs to the field of data processing, specifically referring to a short-term precipitation prediction method based on big data. This scheme specifically extracts near-surface precipitation-related echo information and removes interference from redundant upper-air elevation angle data; using AWS grid precipitation observations as ground truth anchors, it constructs a joint logical calibration rule for radar reflectivity and surface precipitation; it designs differentiated loss weights, introducing strong and weak scaling factors to impose a heavier loss penalty on heavy precipitation samples; it introduces interference suppression coefficients and interference tolerance boundaries to constrain the loss contribution of interference samples; it designs a loss adjustment function to smooth the loss curve and improve the learning sensitivity of heavy precipitation samples; thereby improving the final short-term precipitation prediction effect.

This invention discloses a remote sensing cloud simulation method and apparatus based on a classification-guided diffusion model, belonging to the field of remote sensing image cloud and fog simulation. The method includes: acquiring and preprocessing multi-temporal remote sensing images to construct a dataset containing registered pairs of cloudless and cloudy images; pre-training a feature extraction network based on cloud mask data and performing unsupervised clustering using the cloud features extracted by the network to achieve automated cloud classification; training a pre-constructed cloud simulation diffusion model using the cloudless image and cloud classification results as input, and the corresponding cloudy image as the ground truth supervision signal, to learn the mapping relationship from cloudless to cloudy states; and inputting the target cloudless image and a specified cloud category into the trained cloud simulation diffusion model to generate a simulated cloud-containing remote sensing image. This invention effectively improves the realism of generated clouds and fog, the naturalness of integration with ground features, and the controllability of generating specific cloud categories, providing sufficient and realistic training samples for cloud-related data-driven models.

Techniques and instruments for training a neural network (NN) for interpolation of data are described. In one aspect, sampled signal data is provided to the NN. Each sampled signal data includes a threshold event information and at least one sample taken during a threshold event of an analog signal. Each threshold event information indicates a duration from a time when the sampled signal data is taken to when the analog signal equals a threshold event value. The NN determines predicted threshold event information corresponding to the threshold event of the analog signal for each sampled signal data. NN parameters are adjusted based on errors between each predicted threshold event information and a ground truth threshold event information for each sampled signal data.

The disclosure provides a communication method for localization based on ground truth points. The method includes: receiving ground truth point information related to one or more ground truth points; and obtaining measurement results related to a target based on the ground truth point information. The ground truth point information is used for at least one of sensing the target, positioning the target, or determining a location of the target. Since the GTP information is used as reference, the accuracy of the measurement results related to the target which is determined by the terminal device may be improved.

Some procedures for collecting training data for artificial intelligence of machine learning (AI / ML) positioning may include collecting ground truth information from a user equipment (UE), and collecting radio frequency (RF) fingerprints between a UE and network nodes. An RF fingerprint may be information that characterizes an RF environment. Ground truth information may be radio access technology (RAT)-dependent (e.g., new radio (NR)-based positioning) or RAT-independent. To alleviate the burden on the amount of collected data with absolute ground truth information, semi-supervised techniques may be utilized, where the absolute ground truth location of UEs may be obtained for fewer locations or points. Some examples of the techniques described herein may provide procedures for collecting training data for one or more AI / ML positioning models. Some of the techniques may utilize device-to-device (D2D) positioning measurements, which may be used for training AI / ML models in a semi-supervised fashion.

The present disclosure relates to linking record objects between systems of record based on a comparison of object field-value pairs to a ground truth. A domain name may be identified from an electronic activity. It may be determined that the electronic activity does not match with any first record objects. A second record object including the domain name as a value may be identified. Object field-value pairs of the second record object may be identified. It may be determined that a third record object matches with the second record object. The electronic activity may be matched to the third second record object or a fourth record object. An association between the electronic activity and the third record object or the fourth record object may be stored.

This invention discloses an InSAR interferometric network optimization method based on a multi-factor coherence surrogate model. The method includes: constructing a multi-factor feature dataset containing SAR imagery, baseline data, and soil moisture data of the study area; constructing a candidate interferometric pair set, and randomly selecting a subset of sample interferometric pairs from the candidate set; constructing a coherence surrogate prediction model, which uses the multi-factor feature data corresponding to the sample interferometric pair subset and the ground truth values ​​of sample coherence to train the model parameters; inputting the multi-factor feature data corresponding to each interferometric pair in the candidate set into the trained coherence surrogate prediction model and outputting the predicted coherence of each interferometric pair and the optimized interferometric pair network. This invention improves the stability and accuracy of unwrapping and time-series inversion, while also reducing the difficulty of interferometric processing and coherence calculation under large-scale data, and can support the needs of large-scale, long-term, and near-real-time monitoring.

Certain aspects of the disclosure provide a method for generating an evaluation dataset. The method may include accessing a plurality of vector embeddings associated with a plurality of documents, wherein each vector embedding corresponds to a document; clustering the plurality of vector embeddings into a plurality of clusters; selecting, from each cluster of the plurality of clusters, a subset of vector embeddings of the plurality of vector embeddings according to a sampling strategy; for each respective vector embedding of the subset of vector embeddings: generating a seed question based on content of the portion of the document associated with the respective vector embedding; transforming the seed question into a transformed question of a question type different from the seed question; and associating the transformed question with the respective vector embedding as a ground truth label; and forming an evaluation dataset comprising the transformed questions and their associated ground truth labels.

This disclosure relates to a training method for a multi-label recognition model for speech recognition and a smart home device. The multi-label recognition model includes a shared feature extraction network and multiple branch recognition networks. The shared feature extraction network outputs shared features based on audio data, and the multiple branch recognition networks output multiple recognition labels based on the shared features. The training method includes: acquiring training data, which includes audio data and the ground truth value of at least one of multiple recognition labels associated with the audio data; using the multi-label recognition model to output a predicted value for each of the multiple recognition labels based on the audio data; calculating a branch loss function value between the ground truth value and the predicted value for each of the at least one recognition label; calculating a total loss function value based on the branch loss function value; and updating the parameters of the multi-label recognition model based on the total loss function value.

A method and an apparatus for estimating a vanishing point are provided. The method includes receiving an input image and estimating a vanishing point for the input image, using an artificial intelligence network pre-trained by vanishing point ground truth (GT) data generated based on real image data. Estimating the vanishing point includes estimating a depth map or an optical flow map for the input image, estimating a gradient map for the depth map or the optical flow map, and estimating the vanishing point for the input image based on the gradient map and a predetermined reference gradient map.

Artificial intelligence for time-series data analytics is provided. A first time-series data set is provided to a pre-trained recurrent neural network trained based on a second time-series data set. A prediction of a ground truth state of the first time-series data set is received therefrom. The first time-series data set is provided to a dynamical recurrent neural network trained based on the second time-series data set and the pre-trained recurrent neural network. A noise-reduced prediction of a ground truth system state of the first time-series data set is received therefrom. An estimate of sensor noise is read. The estimate of sensor noise is generated based on the second time-series data set and the pre-trained recurrent neural network. A prediction of a state of the system is generated based on the pre-trained recurrent neural network, the dynamical recurrent neural network, and the estimate of sensor noise.

This invention discloses a self-supervised learning-based atmospheric phase compensation method for InSAR images, belonging to the field of differential interferometry (DI) technology for synthetic aperture radar. This method does not rely on external ground truth data. By establishing constraints for solving the atmospheric phase and integrating them into the network training process, it achieves self-supervised learning for atmospheric phase compensation. The method includes: constructing an atmospheric phase estimation network, which takes a time-series unwrapped phase image and related auxiliary data as input, and outputs the atmospheric phase estimation result or intermediate variables used to calculate the atmospheric phase; constructing differentiated constraint loss terms for reference points with different phase change modes based on the phase component composition of the interferometric phase; embedding physical prior constraints in both network structure and loss terms; combining various loss terms into a total loss function, and completing self-supervised training through iterative optimization to achieve atmospheric phase estimation and compensation.

This invention discloses a hierarchical supervised super-resolution reconstruction method, system, device, and medium for remote sensing images. The method includes acquiring a high-resolution original remote sensing image, preprocessing it to obtain multiple sets of sub-pixel remote sensing images; constructing a sub-pixel remote sensing image reconstruction network, and pre-training it using a low-resolution remote sensing image and a set of sub-pixel remote sensing images as input and ground truth, respectively; constructing an upsampling-free remote sensing image super-resolution network, and optimizing it using multiple sets of sub-pixel images; constructing a multi-level supervised upsampling-free super-resolution network, and training it using an intermediate-resolution remote sensing image and a high-resolution original remote sensing image as ground truth, respectively; and performing super-resolution reconstruction based on the remote sensing image super-resolution reconstruction model, inputting the remote sensing image whose resolution needs to be improved, to obtain a high-resolution remote sensing image. This invention, by employing a hierarchical supervised and upsampling-free super-resolution reconstruction deep network, can improve the quality and detail of super-resolution reconstruction of remote sensing images.

A computer-implemented method for seismic processing includes receiving a seismic training input image, generating, using a first portion of a machine learning model, a first output based at least in part on the seismic training input image, generating, using a second portion of the machine learning model, a second output based at least in part on the seismic training input image, generating a loss function based at least in part on comparing at least two of the first output, a deterministic first label synthetically generated and representing a deterministic ground truth for the first output, the second output, and a non-deterministic second label representing a non-deterministic ground truth for the second output, and refining the first portion, the second portion, or both of the machine learning model based at least in part on the loss function.

The invention relates to a method for training a machine learning model for an autonomous driving function, in particular for behavior prediction and / or behavior planning and / or for tracking one or more vehicles, the method comprising the steps of: providing (S1) a training data set with training data elements, each comprising scenario data (204) as training input data and associated ground truth data; determining (S2) latent feature vectors for the scenario data (204) using a feature embedding unit; clustering (S3) the training data elements based on the latent feature vectors into scenario clusters (202) using a clustering algorithm;Determine (S4) scenario cluster weights by applying the scenario data of the training data elements to the machine learning model to be trained and comparing the output data of the machine learning model thus generated with the ground truth data of the respective training data elements; determine a performance measure and / or a loss function based on the comparisons between the output data and the ground truth data and determine or adjust the scenario cluster weights based on the performance measure and / or the loss function; train (S5) the machine learning model with the training data set sampled taking into account the scenario cluster weights; and deploy (S6) the trained machine learning model.

This disclosure relates to a machine learning method for learning a function configured to predict the next feature in a feature tree. The method includes a dataset providing examples, each example including a graph representing at least a portion of the feature tree. The graph includes nodes, each node representing a feature, and each node is labeled with a label from a set of labels, each label indicating a feature type of a predetermined set. The graph includes edges, each edge connecting nodes and representing parent-child relationships between features represented by the nodes. The graph includes ground truth data indicating the next feature based on one or more selected features, each selected feature corresponding to a node in the graph. The method also includes training the function to take the graph representing at least a portion of the feature tree as input and output a prediction of one or more next features in the feature tree.

This application provides a facial skeleton rigging method, apparatus, electronic device, and storage medium. The facial skeleton rigging method includes: binding facial bones to facial mesh data in the initial state of the character to be constructed, obtaining a first binding result; loading a historical skeleton controller and first controller parameters of the historical skeleton controller; calculating the facial bone position using the first controller parameters, obtaining the vertex position of the first facial mesh; and optimizing the first binding result based on the vertex position of the first facial mesh, the facial mesh data, and the first controller parameters to obtain a second binding result. Calculating the facial bone position using controller parameters and optimizing the first binding result using the vertex position of the facial mesh and the facial mesh data as ground truth significantly improves the rigging effect of facial skeleton rigging.

This invention relates to a two-stage super-resolution method for MRI images using high-quality codebook prior information. The method includes: a) data preprocessing, undersampling the fully sampled K-space data; b) designing a one-stage codebook network, pre-trained using ground truth images to obtain high-quality codebook information; d) designing a two-stage codebook-prior super-resolution network model that integrates prior information; e) supervised training of the model using ground truth images; and f) selecting the optimal model based on a validation set, inputting test data into the model to obtain the super-resolution MRI image. Compared with existing technologies, this invention introduces high-quality codebook prior information during the super-resolution process, effectively solving the problem of poor restoration of details and texture structure in super-resolution images. Better restoration of detailed texture information aids in clinical diagnosis and has promising application prospects.

A method for generating a trained machine learning model trained on multiple segregated data sources includes generating a first dataset by transforming a first source dataset by generating an embedded representation of the first source dataset and adding privacy parameters. The method includes generating a second dataset by transforming a second source dataset by generating an embedded representation of the second source dataset and adding privacy parameters, generating a combined dataset that includes the first dataset and a ground truth dataset from a first segregated data environment combined with the second dataset from a second segregated data environment (e.g., within a trusted research environment or a secure processing environment). The method includes training a machine learning model with training data that includes a subset of the combined dataset, in which the model parameters of the trained machine learning model are stored in a storage device.