A method of an electronic device including obtaining a low-resolution input image by down-sampling a high-resolution input image; obtaining a low-resolution output image by performing image quality processing on the low-resolution input image; obtaining a low-resolution model from a conversion relationship between the low-resolution input image prior to the image quality processing being performed and the low-resolution output image subsequent to the image quality processing being performed; performing up-sampling of the low-resolution model; obtaining a high-resolution model by modifying the up-sampled low-resolution model, based on a difference between the high-resolution input image and the low-resolution input image; and obtaining a high-resolution output image from the high-resolution input image, by applying the high-resolution model to the high-resolution input image.

Examples described herein provide a method that includes performing a first scan of an object to generate first scan data. The method further includes detecting a defect on a surface of the object by analyzing the first scan data to identify a region of interest containing the defect by comparing the first scan data to reference scan data. The method further includes performing a second scan of the region of interest containing the defect to generate second scan data, the second scan data being higher resolution scan data than the first scan data. The method further includes combining the first scan data and the second scan data to generate a point cloud of the object.

A method includes obtaining (such as accessing, receiving, acquiring, etc.), using at least one processor of an electronic device, a machine learning model trained to process input data and generate output data over at least one range of values associated with one or more control variables. The method also includes providing, using the at least one processor, specified input data to the machine learning model and providing, using the at least one processor, one or more specified values of the one or more control variables to the machine learning model. The one or more specified values of the one or more control variables are within the at least one range of values. The method further includes performing inferencing using the machine learning model to process the specified input data and generate specified output data. The inferencing is controlled based on the one or more specified values of the control variable(s).

An apparatus includes a sensor having an array of detectors. The sensor is configured to assign multiple detectors to a detector group corresponding to a block pixel. The sensor is also configured, for each frame of a set of frames, to apply a specified one of a set of mask patterns in order to select outputs of the detectors in the detector group and aggregate the selected outputs of the detectors in the detector group to determine pixel information for the block pixel. The apparatus also includes at least one processor configured to generate the frames using the pixel information for the block pixel, and upscale the portion of the at least one of the frames using the set of mask patterns to identify native pixels within the block pixel.

An apparatus includes a sensor having an array of detectors. The sensor is configured to assign multiple detectors to a detector group corresponding to a block pixel. The sensor is also configured, for each frame of a set of frames, to apply a specified one of a set of mask patterns in order to select outputs of the detectors in the detector group and aggregate the selected outputs of the detectors in the detector group to determine pixel information for the block pixel. The apparatus also includes at least one processor configured to generate the frames using the pixel information for the block pixel, and upscale the portion of the at least one of the frames using the set of mask patterns to identify native pixels within the block pixel.

One embodiment provides a method generating a first image crop and a second image crop randomly extracted from a low-quality image and a high-quality image, respectively. The method further comprises comparing the first image crop and the second image crop using a plurality of loss functions including pixel-wise loss to calculate losses, and optimizing a model trained to estimate a realistic scale-independent blur kernel of a low-resolution (LR) blurred image by minimizing the losses.

The apparatus for an X-ray data generation according to an embodiment of the inventive concept includes a processor that receives 3D data to generate output data and a buffer, and the processor includes an extraction unit that extracts raw object data from the 3D data and projects the raw object data onto a 2D plane to generate first object data, an augmentation unit that performs data augmentation on the first object data to generate second object data, a composition unit that synthesizes the second object data and background data to generate composite data, and a post-processing unit that performs post-processing on the composite data to generate the output data, and the buffer stores a plurality of parameters related to generation of the first object data, the second object data, the composite data, and the output data.

The apparatus for an X-ray data generation according to an embodiment of the inventive concept includes a processor that receives 3D data to generate output data and a buffer, and the processor includes an extraction unit that extracts raw object data from the 3D data and projects the raw object data onto a 2D plane to generate first object data, an augmentation unit that performs data augmentation on the first object data to generate second object data, a composition unit that synthesizes the second object data and background data to generate composite data, and a post-processing unit that performs post-processing on the composite data to generate the output data, and the buffer stores a plurality of parameters related to generation of the first object data, the second object data, the composite data, and the output data.

In a laser cataract procedure that also corrects for astigmatism, an iris registration method compares an iris image of a patient's eye taken when the eye is not docked to a patient interface device with an iris image of the same eye that is docked to the patient interface, to calculate a rotation angle between the two images. The astigmatism axis of the eye is measured when the eye is not docked, and the measured axis is rotated by the calculated rotation angle to obtain a rotated astigmatism axis relative to the iris image of the docked eye. The laser cataract procedure is performed based on the rotated astigmatism axis. The rotation angle is calculated by optimizing a transformation that transforms the undocked iris image to match the docked iris image, where the transformation includes a dilation factor that accounts for different pupil dilation of the two iris images.

A method, apparatus, and computer program product are provided for training a neural network or providing a pre-trained neural network with the weight-updates being compressible using at least a weight-update compression loss function and/or task loss function. The weight-update compression loss function can comprise a weight-update vector defined as a latest weight vector minus an initial weight vector before training. A pre-trained neural network can be compressed by pruning one or more small-valued weights. The training of the neural network can consider the compressibility of the neural network, for instance, using a compression loss function, such as a task loss and/or a weight-update compression loss. The compressed neural network can be applied within a decoding loop of an encoder side or in a post-processing stage, as well as at a decoder side.