Disclosed is a method for recognizing a marine object based on hyperspectral data including collecting target hyperspectral data; preprocessing the target hyperspectral data; and detecting and identifying an object included in the target hyperspectral data based on a marine object detection and identification model, trained through learning of the detection and identification of the marine object. According to the present invention, the preprocessing and processing of the hyperspectral data collected in real time according to a communication state may be performed in the sky or on the ground.

Training deep neural networks requires a large amount of labeled training data. Conventionally, labeled training data is generated by gathering real images that are manually labelled which is very time-consuming. Instead of manually labelling a training dataset, domain randomization technique is used generate training data that is automatically labeled. The generated training data may be used to train neural networks for object detection and segmentation (labelling) tasks. In an embodiment, the generated training data includes synthetic input images generated by rendering three-dimensional (3D) objects of interest in a 3D scene. In an embodiment, the generated training data includes synthetic input images generated by rendering 3D objects of interest on a 2D background image. The 3D objects of interest are objects that a neural network is trained to detect and/or label.

A method for training a neural network, including: determining a neural network; training the neural network at a first learning rate according to a first optimization mode, where the first learning rate is updated each time the neural network is trained; mapping the first learning rate of the first optimization mode to a second learning rate of a second optimization mode in the same vector space; determining the second learning rate satisfies a preset update condition; and continuing to train the neural network at the second learning rate according to the second optimization mode.

Embodiments of the present disclosure relate to a method, an electronic device, and a computer program product for data augmentation. The method includes: generating a group of candidate images based on a target image by using a thermodynamic genetic algorithm (TDGA) model, the TDGA model being configured to apply one or more operations of a set of predetermined image processing operations during each evolution process; and determining multiple augmented images from the group of candidate images based on free energy of the group of candidate images, the multiple augmented images being determined as belonging to the same classification with the target image. In this way, data augmentation can be efficiently implemented by a thermodynamic genetic algorithm.

A computing entity is described that obtains at least one inspection image of an at least partially fabricated product and causes the at least one inspection image to be processed by a product inspection engine. The product inspection engine includes a machine learning-trained model. The computing entity obtains an inspection result determined based on the processing of the at least one inspection image by the product inspection engine; identifies one or more training images stored in an image database based at least in part on the at least one inspection image; associates automatically generated labeling data with the one or more training images based at least in part on the inspection result determined by the processing of the at least one inspection image; and causes training of the product inspection engine using the one or more training images and the associated labeling data.

Embodiments of the present disclosure relate to a method, an electronic device, and a computer program product for generating an avatar. The method includes generating an indication of correlation among image information, audio information, and text information of a video. The method may further include generating, based on the indication of the correlation, a first feature set and a second feature set representing features of a target object in the video, wherein the first feature set represents invariant features of the target object in the video, and the second feature set represents equivariant features of the target object in the video. The method may further include generating the avatar based on the first feature set and the second feature set. With this method, the generated avatar can be made more accurate and vivid with a better effect, while also reducing data annotation cost, improving operation efficiency, and enhancing user experience.

Systems, methods, and computer-readable media are disclosed for dynamically adjusting a configuration of a pre-processor and/or a post-processor of a machine learning system. In one aspect, a machine learning system can receive raw data at a pre-processor where the pre-processor being configured to generate pre-processed data, train a machine learning model based on the pre-processed data to generate output data, process the output data at a post-processor to generate inference data, and adjust, by a controller, configuration of one or a combination of the pre-processor and the post-processor based on the inference data.

A method and apparatus is provided for generating an augmented sample set for enriching a first training dataset for training a model. The method comprises: using data augmentation and corresponding labeling or using label augmentation to add a first augmented sample set to the first training dataset, wherein the data augmentation and corresponding labeling, or the label augmentation purposely puts a first distinguishing characteristic of a first part-of-interest or an associated label into the first training dataset to cause the first distinguishing characteristic of the first part-of-interest to be emphasized to enable the model to learn a generalizable principle of the first distinguishing characteristic, wherein the first distinguishing characteristic is for differentiating the first part-of-interest from a second part-of-interest. Methods for training a model, using a model to differentiate part-of-interests and using a model to infer a dataset are also provided.

A system and a method of performing face recognition may include: receiving a first facial image, depicting a first face, and a second facial image depicting a second face; applying an ML model on the first image, to produce a first representation vector, and applying the ML model on the second image to produce a second representation vector; comparing the first representation vector and the second representation vector; and associating the first face with the second face based on the comparison, where the ML model is trained to produce the representation vectors from the facial images, based on regions in the facial images that correspond to distinctiveness scores that are beneath a distinctiveness threshold.

A method of analyzing a product includes performing an anomaly detection on a received image using an autoencoder, wherein the autoencoder includes at least one first neural network trained based on a first set of training images, and the first set of training images includes a plurality of training images each showing a corresponding defect-free product; determining, using a binary classifier, whether or not a defect is present based on a result of the anomaly detection; performing defect detection on the received image using a defect detector, wherein the defect detector includes a third neural network trained based on a one third set of training images, and the third set of training images includes a plurality of training images each showing a corresponding defective product; and evaluating a result based on a weighting of the results of the anomaly detection, the defect detection, and the binary classifier.