A method for training a neural network includes receiving a plurality of images and, for each individual image of the plurality of images, generating a training triplet including a subset of the individual image, a subset of a transformed image, and a homography based on the subset of the individual image and the subset of the transformed image. The method also includes, for each individual image, generating, by the neural network, an estimated homography based on the subset of the individual image and the subset of the transformed image, comparing the estimated homography to the homography, and modifying the neural network based on the comparison.

Image processing method and device for a spliced panel and a spliced panel are provided. The image processing method includes calculating a horizontal stretch coefficient of the spliced panel according to a resolution of the plurality of display units, a sum of horizontal spliced-gap widths of the spliced panel and a horizontal pixel pitch of the plurality of display units; calculating a vertical stretch coefficient of the spliced panel according to the resolution of the plurality of display units, a sum of vertical spliced-gap widths of the spliced panel and a vertical pixel pitch of the plurality of display units; stretching an original image to be displayed by the spliced panel according to the horizontal stretch coefficient and the vertical stretch coefficient to acquire a target image; and controlling display of the plurality of display units of the spliced panel according to the target image.

Aspects presented herein relate to methods and devices for display processing including an apparatus, e.g., a DPU. The apparatus may receive at least one input image for a scaling operation, the at least one input image being associated with one or more scanning windows, each of the scanning windows including a plurality of pixels. The apparatus may also detect one or more features in the plurality of pixels in each of the one or more scanning windows. Further, the apparatus may adjust an amount of the plurality of pixels in each of the scanning windows for each of the detected features. The apparatus may also combine the adjusted amount of the plurality of pixels for each of the detected one or more features into a plurality of output pixels. The apparatus may also process each of the plurality of output pixels into at least one output image.

Video stream data is selectively scaled so that sections within regions of interest (ROI) maintain high resolution while areas not within the region of interest are down-scaled to reduce bandwidth cost of transmission. A low compression encoder compresses sections of a video frame corresponding to one or more ROI without motion search or prediction mode decision to generate low-compression section data. The video frame is downscaled and a high compression encoder compresses the resulting downscaled video frame with prediction mode decision to generate high-compression frame data.

An apparatus and method are described for a non-uniform rasterizer. For example, one embodiment of an apparatus comprises: a graphics processor to process graphics data and render images using the graphics data; and a non-uniform rasterizer within the graphics processor to determine different resolutions to be used for different regions of an image, the non-uniform rasterizer to receive a plurality of polygons to be rasterized and to responsively rasterize the polygons in accordance with the different resolutions.

Systems and methods are provided for performing temporally consistent depth map generation by implementing acts of obtaining a first stereo pair of images of a scene associated with a first timepoint and a first pose, generating a first depth map of the scene based on the first stereo pair of images, obtaining a second stereo pair of images of the scene associated with at a second timepoint and a second pose, generating a reprojected first depth map by reprojecting the first depth map to align the first depth map with the second stereo pair of images, and generating a second depth map that corresponds to the second stereo pair of images using the reprojected first depth map.

The disclosure discloses a reconfigurable hardware acceleration method and system for Gaussian pyramid construction and belongs to the field of hardware accelerator design. The system provided by the disclosure includes a static random access memory (SRAM) bank, a first in first out (FIFO) group, a switch network, a shift register array, an adder tree module, a demultiplexer, a reconfigurable PE array, and a Gaussian difference module. In the disclosure, according to the requirements of different scenarios and different tasks for the system, reconfigurable PE array resources can be configured to realize convolution calculations of different scales. The disclosure includes methods of fast and slow dual clock domain design, dynamic edge padding design, and input image partial sum reusing design.

An image processing method includes: according to position coordinates of any interpolation pixel in a target image, determining position coordinate of the interpolation pixel in an original image; calculating a two-dimensional image entropy of an (n×n) neighborhood of the interpolation pixel in the original image; when it is greater than or equal to a preset entropy threshold value, calculating a pixel value of the interpolation pixel based on all original pixels; when it is less than the preset entropy threshold, calculating gradient values in at least two edge directions within the threshold neighborhood and determining whether there is a strong-edge direction; if so, calculating the pixel value of the interpolation pixel based on a plurality of original pixels in the strong-edge direction; if not, calculating the pixel value of the interpolation pixel based on all the original pixels.

A system is adapted to perform an image processing method. The processing method includes: obtaining input image data, a first training result, a second training result, and an interpolation lookup table; segmenting the input image data into a plurality of feature blocks according to a total quantity of area interpolations; establishing a position mapping relationship to record the feature blocks corresponding to positions of all of the area interpolations; assigning corresponding area interpolations to the feature blocks according to the position mapping relationship; obtaining an interpolation parameter for each of the feature blocks according to the first training result, the second training result, and the area interpolation; performing block convolution on each of the interpolation parameters and the corresponding feature block to obtain an output feature result; and obtaining an output image by combining the output feature results according to the position mapping relationship.

In one aspect, an example method for generating a candidate image for use as backdrop imagery for a graphical user interface is disclosed. The method includes receiving a raw image and determining an edge image from the raw image using edge detection. The method also includes identifying a candidate region of interest (ROI) in the raw image based on the candidate ROI enclosing a portion of the edge image having edge densities exceeding a threshold edge density. The method also includes manipulating the raw image relative to a backdrop imagery canvas for a graphical user interface based on a location of the candidate ROI within the raw image. The method also includes generating, based on the manipulating, a set of candidate backdrop images in which at least a portion of the candidate ROI occupies a preselected area of the backdrop imagery canvas, and storing the set of candidate backdrop images.